Data processing system

Abstract

The present application relates to a system for realizing personnel evaluation of managers fairly and transparently, wherein a data collection unit automatically collects various personnel-related data such as turnover rate, willingness to move rate, promotion performance, performance evaluation, anonymous feedback, and the like, a generative AI analysis unit uses the data to analyze turnover rate or willingness to move rate, promotion rate, bias detection, trust score, improvement points, and the like from multiple angles, and an evaluation report generation unit automatically generates and distributes an evaluation report for each manager. Thus, the subjective and individualized evaluation of the past is eliminated, and evaluation based on objective and quantitative indicators, utilization of feedback, and a highly transparent evaluation process are realized, which can improve the trustworthiness and performance of the entire organization.

Description

Technical Field

[0001] The technology disclosed herein relates to a system. Background Technology

[0002] Patent Document 1 discloses a personalized chatbot control method executed by at least one processor, the method comprising: receiving user speech; adding the user speech to a prompt containing instructions related to a chatbot role; encoding the prompt; and inputting the encoded prompt into a language model to generate chatbot speech in response to the user speech.

[0003] Patent document 1: Japanese Patent Application Publication No. 2022-180282.

[0004] The problem this invention aims to address is that, in the personnel evaluation system for managers within organizations, there have historically been a complex interplay of issues, including superiors' subjective biases or preferences, individualized judgments, opaque evaluation standards, insufficient utilization of feedback from subordinates, the infiltration of prejudice, neglect of subordinates' career development support performance, and increased turnover or relocation intentions. These issues have led to serious problems such as decreased employee motivation or talent loss, unfair evaluations, a superficial organizational culture, and obstacles to building healthy trust relationships. Specifically, the following issues exist: the turnover, job-hopping, and relocation intentions of managers' subordinates, or their promotion performance, are not properly evaluated. Even when subordinates provide anonymous and frank feedback, its content is not properly analyzed and utilized, making it impossible to objectively assess the manager's management capabilities or trustworthiness. Furthermore, due to the opaque nature of the evaluation process, evaluation standards or results are not fully disclosed to relevant personnel, significantly compromising the transparency and credibility of the evaluation. To address these issues, it is necessary to collect a wide range of personnel data or anonymous feedback about managers' subordinates in a comprehensive and automated manner. Through advanced analysis based on generative AI, quantitative and objective data such as turnover rate, transfer intention rate, promotion rate, bias, trust score, and areas for improvement can be visualized and provided to relevant personnel as evaluation reports on a regular or real-time basis. This will enable an evaluation system that is both fair and transparent, thereby improving trust and performance throughout the organization. Summary of the Invention

[0005] The means to solve the problem involved in this invention is to construct a system comprising a data collection department, a generative AI analysis department, and an evaluation report generation department for the fair and transparent evaluation of managers. First, in the data collection department, various personnel-related data, such as employee turnover, job-hopping, transfer intentions, promotion performance, performance evaluations, potential scores, and anonymous feedback, are automatically and comprehensively collected from multiple databases or feedback systems both inside and outside the organization. Next, in the generative AI analysis department, using the various data obtained by the data collection department, the turnover rate, transfer intention rate, and promotion rate of each manager's employees are quantitatively calculated, and future promotion possibilities are predicted based on the performance or potential of the employees. Furthermore, the distribution of evaluation scores given by managers or the evaluation tendency of each attribute is analyzed to detect any subjective bias. Finally, natural language analysis of anonymous feedback is used to extract the manager's management style, trust score, and areas for improvement. Finally, in the evaluation report generation department, an evaluation report is automatically generated for each manager, including turnover rate, transfer intention rate, promotion rate, bias detection results, trust score, and list of improvement points obtained by the generative AI analysis department. The report is then distributed regularly or in real time to the manager, their supervisor, and the human resources department, thereby ensuring the transparency and fairness of the evaluation process and improving the trust and performance of the entire organization. Attached Figure Description

[0006] Figure 1 This is a conceptual diagram illustrating an example of the configuration of a data processing system according to the first embodiment.

[0007] Figure 2 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and smart device according to the first embodiment.

[0008] Figure 3 This is a conceptual diagram illustrating an example of the data processing system configuration in the second embodiment.

[0009] Figure 4 This is a conceptual diagram illustrating an example of the main functions of the data processing device and smart glasses according to the second embodiment.

[0010] Figure 5 This is a conceptual diagram illustrating an example of the data processing system configuration in the third embodiment.

[0011] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0012] Figure 7 This is a conceptual diagram illustrating an example of the data processing system configuration in the fourth embodiment.

[0013] Figure 8This is a conceptual diagram illustrating an example of the functions of the main parts of the data processing device and robot according to the fourth embodiment.

[0014] Figure 9 It represents an emotion graph that maps multiple emotions.

[0015] Figure 10 It represents an emotion graph that maps multiple emotions.

[0016] Explanation of reference numerals in the attached figures Data processing systems 10, 210, 310, and 410 12 Data processing device 14 Smart devices 214 Smart Glasses 314 Head-mounted terminal 414 Robot. Detailed Implementation

[0017] Hereinafter, an example of an implementation of the system involved in this disclosure will be described with reference to the accompanying drawings.

[0018] First, let's explain the terms used in the following description.

[0019] In the following embodiments, the processor (hereinafter referred to as "processor") can be a single computing device or a combination of multiple computing devices. Furthermore, a processor can be a single computing device or a combination of multiple computing devices. Examples of computing devices include CPU (Central Processing Unit), GPU (Graphics Processing Unit), GPGPU (General-Purpose computing on Graphics Processing Units), APU (Accelerated Processing Unit), etc.

[0020] In the following implementation, the labeled RAM (Random Access Memory) is a memory that temporarily stores information and is used by the processor as working memory.

[0021] In the following embodiments, the labeled memory is one or more non-volatile storage devices used to store various programs and various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disk (e.g., hard disk) or magnetic tape, etc.

[0022] In the following implementation, the labeled Communication I / F (Interface) is an interface that includes a communication processor and an antenna, etc. The Communication I / F is responsible for communication between multiple computers. Examples of communication standards applicable to the Communication I / F include wireless communication standards such as 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0023] In the following implementation, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. Furthermore, in this specification, when "and / or" connects more than three items, the same approach as "A and / or B" applies.

[0024] [First Implementation] Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0025] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. An example of the data processing device 12 is a server.

[0026] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0027] The smart device 14 includes a computer 36, a receiver 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the receiver 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0028] The receiving device 38 includes a touchscreen 38A and a microphone 38B, etc., for receiving user input. The touchscreen 38A receives user input generated by contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input generated by sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing user input received through the touchscreen 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data representing the user input.

[0029] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting data in a form perceptible to the user 20 (e.g., sound and / or text). The display 40A displays visual information such as text and images according to instructions from the processor 46. The speaker 40B outputs sound according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0030] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0031] Figure 2 An example of the main functions of the data processing device 12 and the smart device 14 is shown.

[0032] like Figure 2 As shown, in the data processing apparatus 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the memory 32. The specific processing program 56 is an example of a "program" as understood in this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 based on the specific processing program 56 executed on the RAM 30.

[0033] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0034] In the smart device 14, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The receive / output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart device 14 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0035] In addition, other devices besides the data processing device 12 may also have a data generation model 58.

[0036] For example, a server device (e.g., a generation server) may have a data generation model 58.

[0037] In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with a server device that has the data generation model 58. Furthermore, the data processing device 12 can be a server device or a user-owned terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing performed by the data processing system 10 of the first embodiment will be described.

[0038] (Example 1) The specific processing flow in Example 1 will be described below. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. Furthermore, the data processing device 12 is referred to as the "server" and the smart device 14 is referred to as the "terminal".

[0039] (Implementation methods for carrying out the present invention) The embodiments of the present invention will be described in more detail below. The personnel evaluation system of the present invention adopts a client-server architecture consisting of a server and a terminal. On the server side, a data collection unit, a generative AI analysis unit, an evaluation report generation unit, a database management unit, and an alarm notification unit are installed. On the terminal side, a user interface unit, a feedback input unit, a report viewing unit, and an evaluation interview support unit are installed. Specifically, this system operates in a distributed computing environment. The server consists of multiple computing nodes. The server has a parallel computing processor for high-speed execution of large-scale language models and deep learning models. The terminal is an information processing terminal that executes a web browser or a dedicated application. This system uses an encryption protocol in communication between the client and server. The data collection unit uses asynchronous communication to acquire data from multiple external systems. The generative AI analysis unit converts the acquired data into multidimensional tensors. The generative AI analysis unit uses the converted tensors as input to perform inference processing. The evaluation report generation unit formats the inference results into structured data. The database management unit uses a distributed relational database or a non-relational database to persist the data. The alarm notification unit adopts a real-time event-driven architecture. The user interface unit renders dynamic network components. This feedback input section receives text input from users. This report reading section visualizes structured data as charts. This assessment interview support section establishes real-time, two-way communication sessions. Through this structure, the system processes massive amounts of personnel data with low latency. This system dramatically improves the computational efficiency of the evaluation process. This system ensures the objectivity of personnel evaluations in large-scale organizations. As a specific application area, this system is used for global talent management in multinational corporations with tens of thousands of employees. This system provides consistent evaluation standards across different time zones and language environments. This system reduces communication load during peak periods through dynamic allocation of hardware resources. This system balances scalability and high availability. This system completely digitizes the previously subjective evaluation process performed by humans. This system supports data-driven decision-making. This system directly contributes to improving the productivity of the entire organization.

[0040] First, the data collection department operates on a server, automatically and regularly collecting data from multiple internal and external systems and databases within the organization, including HR management, business management, project management, employee questionnaires, anonymous feedback, attendance management, company social networking, email, and training management. This is achieved through API integration, batch processing, and file import. For example, it collects comprehensive information about employees' subordinates from multiple angles, such as departure date, transfer request date, any job-hopping activity, past promotion history, performance evaluation scores, potential scores, anonymous feedback, 360-degree feedback results, attendance data (lateness, early departure, overtime, paid leave status), project participation history, submission of business deliverables, frequency and content of posts on company social networking, number of emails sent and received, training experience, qualification status, records of self-motivation activities, health diagnosis results, and stress test results. For instance, it can obtain detailed data such as whether an employee has requested a transfer twice in the past year, participated in three training sessions, served as a project leader once, and averaged 30 posts per month on company social networking. Specifically, this data collection unit implements dedicated connector modules for multiple heterogeneous data sources. It communicates with external systems using a stateless communication protocol. The unit normalizes the acquired raw data into an intermediate data format. It analyzes unstructured and semi-structured data, including natural language text, numerical data, and timestamps. It performs preprocessing such as imputation of missing values ​​and removal of outliers. It transforms the preprocessed data into multidimensional feature vectors. For example, it encodes employee action records as time-series numerical arrays. It maps text data to a high-dimensional vector space using word embedding methods. It integrates these vector data to generate a single input tensor. This generated input tensor is forwarded to the subsequent generative AI analysis unit. Through this processing, the system can handle massive amounts of data with different formats using a unified computational model. The system significantly reduces the computational cost required for data integration. It surpasses the limits of manual data collection, achieving real-time and comprehensive data processing. As a specific application area, this system is used for talent allocation optimization in large enterprises with complex matrix organizations. This system quantitatively captures subtle changes in employee behavior. It facilitates the early detection of turnover risks and the discovery of hidden talent. This data collection department uses distributed message queues to process data asynchronously. This data collection department enhances the system's fault tolerance. Even with increasing data volumes, this data collection department maintains processing speed through horizontal scaling. This data collection department combines periodic batch processing with real-time stream processing. This data collection department consistently maintains the highest levels of data freshness and accuracy.

[0041] The Database Management Department organizes and manages this diverse data according to time series, enabling retrieval, extraction, and aggregation from multiple perspectives, such as individual managers, employees, departments, and projects. For example, it builds a data foundation for detailed analysis, such as the turnover rate of employees under a specific manager over the past three years, annual comparisons of transfer intention rates in a department, and the distribution of performance evaluations for members in a specific project. Specifically, the Database Management Department implements high-availability data storage built on a distributed file system. It constructs index structures specifically for time-series data. It maintains tree-structured or hash indexes in memory for high-speed processing of multidimensional retrieval queries. It normalizes data using manager and employee identifiers as primary keys. It defines relationships using department codes and project identifiers as foreign keys. It logs transactions every time data is added or updated. It ensures data consistency and durability. It uses compression algorithms to push past historical data to cold storage. It caches frequently accessed, up-to-date data in memory. This system achieves millisecond-level response times even for massive datasets at the petabyte level. It evenly distributes the computational load of complex aggregation processing across distributed nodes. It eliminates database bottlenecks and improves the overall system's processing capacity. In a specific application area, this system is used in large, long-established enterprises with decades of personnel data to develop long-term talent development plans. It enables the instantaneous retrieval of vast past success and failure cases. It provides the foundation for building sophisticated data-driven predictive models. The database management department implements encrypted data storage as a standard practice. It minimizes the risk of personal information leakage. It strictly manages each user's viewing permissions using access control lists. It automatically generates audit logs to meet compliance requirements.

[0042] The generative AI analytics department runs on a server and performs diverse AI analyses using data stored in the database management department. Generative AI, for example, combines large-scale language models, gradient boosting models, random forests, neural networks, natural language processing models, clustering models, anomaly detection models, etc. First, to calculate the turnover rate and transfer intention rate of each manager's subordinates, for example, by summarizing the number of employees who left and the number who expressed a transfer intention in the past year, and dividing by the total number of subordinates, a quantitative indicator is obtained. For example, if a manager has 20 subordinates, and 3 have left and 2 have expressed a transfer intention, the turnover rate is 15% and the transfer intention rate is 10%. Specifically, this generative AI analytics department employs an ensemble learning architecture that integrates multiple machine learning models. This generative AI analytics department receives a multidimensional tensor containing employee action history and text data as input data. This input tensor is, for example, a 640-dimensional vector formed by combining a 512-dimensional text embedding vector and a 128-dimensional numerical feature vector. As a specific example of the input tensor, it contains a time-series array of past evaluation scores and an array of sentiment scores extracted from company communication tools. This generative AI analysis unit feeds these input tensors into a neural network equipped with a multilayer perceptron and a self-attention mechanism. This neural network learns long-term dependencies within the time-series data. This neural network is pre-trained using a cross-entropy error function. This neural network optimizes the weight parameters using gradient descent. This generative AI analysis unit outputs a continuous score ranging from 0 to 1, representing the probability of employee turnover and the probability of employee transfer intention, as the inference result. As a specific example of the output score, values ​​such as a turnover probability of 0.85 and a transfer intention probability of 0.62 are calculated. This generative AI analysis unit further aggregates these scores for each manager to calculate the organizational unit's risk index. This generative AI analysis unit sends this risk index to the subsequent evaluation report generation department and the alarm notification department. In this subsequent processing, the system automatically triggers an alarm if the risk index exceeds a predetermined threshold. Through this processing, the system identifies complex correlations in high-dimensional space that might be missed in simple statistical calculations. This system significantly improves prediction accuracy. This system enables objective risk assessment that does not rely on human intuition. In specific application areas, it is used to formulate strategies to prevent employee turnover in rapidly growing emerging companies and companies undergoing large-scale organizational restructuring. This system makes data-driven, proactive personnel management possible. The generative AI analysis unit generates attention maps that visualize the model's reasoning process. The generative AI analysis unit provides the basis for predictions in a human-understandable format.

[0043] Regarding subordinates' promotion performance, the AI ​​calculates the number of employees promoted and the promotion rate over the past two years, and further analyzes the performance evaluation scores, project management experience, and self-help activity records of promoted employees to evaluate the professional support performance of managers from multiple perspectives. For example, if five subordinates are promoted, three of whom have experience as project managers and two have obtained qualifications, the AI ​​determines that they "actively provided professional support." Specifically, this generative AI analysis unit executes a dedicated classification algorithm for evaluating professional support performance. The unit receives structured data as input, including subordinates' promotion history, performance evaluation score trends, project participation history, and qualification acquisition history. This structured data is normalized to generate feature vectors for each employee. Specific examples of these feature vectors include integer values ​​representing the number of promotions over the past two years, floating-point numbers representing the average performance evaluation score, and binary data representing whether or not the employee has management experience. This feature vector is then input into a support vector machine or gradient boosting decision tree. The model undergoes supervised learning using data from past high-performing managers as labels for correct answers. This model classifies the input feature vector in a multidimensional feature space. It calculates a score from 0 to 100 representing a manager's career support ability as output. Specific examples of this output score include a career support score of 85 and a development ability score of 92. The generative AI analysis department forwards this output score to the subsequent evaluation report generation department. In this subsequent processing, the system automatically adjusts the manager's overall evaluation based on this score. Through this process, the system not only quantitatively evaluates the apparent promotion rate but also the underlying development process. This system visualizes the manager's true management ability. This system significantly improves the fairness and persuasiveness of the evaluation. As a specific application area, this system is used for the selection of next-generation leaders and the suitability assessment of management positions. This system accurately identifies talent that drives the organization's long-term growth. The generative AI analysis department calculates the importance of feature quantities, identifying which development actions contribute most to promotion. The generative AI analysis department provides this importance information as feedback to managers. The generative AI analysis department promotes the sharing of development know-how throughout the organization.

[0044] Generative AI uses employee performance evaluation scores, potential scores, business results, project management experience, self-motivation activity history, training history, posts on company social media, health diagnoses, and stress test results as inputs to predict the likelihood of promotion within the next six months. For example, for employees with high performance evaluations, active self-motivation activities, and good stress test results, the AI ​​determines them to have a "high likelihood of promotion," which is reflected in the evaluation of management's career support performance. Specifically, this generative AI analysis department executes a time-series forecasting model to predict employees' future growth. This generative AI analysis department obtains employee performance evaluation scores, potential scores, business results, project management experience, self-motivation activity history, training history, posts on company communication tools, health diagnoses, and stress test results as input data. This generative AI analysis department constructs a multivariate time-series tensor that arranges this data along a time axis. As a specific example of this input tensor, it includes an array of monthly evaluation scores for the past 36 months and a numerical array representing the progression of learning time each month. This generative AI analysis unit inputs the time-series tensor into a Long Short-Term Memory (LSTM) network or a one-dimensional convolutional neural network. The network is trained to predict future changes from past data patterns. The network updates its weights through backpropagation to minimize the mean squared error. The generative AI analysis unit generates a probability distribution representing the probability of promotion within the next six months as output. As a specific example of this output data, predicted values ​​such as a 78% probability of promotion within six months and a 95% probability of promotion within one year are calculated. The generative AI analysis unit saves this prediction result to the subsequent database management department and sends it to the evaluation report generation department. In this subsequent processing, the system uses the prediction result to automatically recommend the best training plan. Through this processing, the system not only scientifically evaluates past performance but also scientifically evaluates future potential. The system maximizes the return on investment in talent development. The system directly helps improve employee motivation. As a specific application area, this system is applied to the design of career paths for highly specialized technical and research positions. The system dynamically generates the best career plan corresponding to an individual's adaptability. The generative AI analysis unit calculates the predicted uncertainty as a confidence interval. This generative AI analytics department supports the development of flexible personnel plans that take risk into account.

[0045] Regarding the detection of bias in managers' evaluations of subordinates, the AI ​​uses statistical and clustering methods. For example, it detects bias and calculates a bias score when the distribution of evaluation scores is extremely skewed, or when there is a bias towards specific attributes (gender, age, job title, department of origin, etc.). For instance, if a manager consistently gives low evaluations to female employees or high evaluations to employees from a specific department, the AI ​​determines that there is a "risk of bias" and records the specific evaluation bias in the report. Specifically, this generative AI analysis department executes anomaly detection algorithms to ensure the fairness of evaluations. This generative AI analysis department obtains a list of evaluation scores assigned by managers and attribute data (gender, age, job title, department of origin, etc.) of the evaluated individuals as input data. As a specific example of this input data, it includes an array of 500 evaluation scores assigned by a specific manager over the past 3 years and attribute vectors of employees corresponding to each score. This generative AI analysis department uses this input data to perform clustering processing in a multidimensional space. This generative AI analysis department uses K-means and density-based spatial clustering algorithms to form groups of evaluation patterns. This generative AI analysis unit calculates the statistical correlation between the clusters formed and the attribute data. When an evaluation score is significantly biased relative to a specific attribute, this unit calculates the bias. The unit generates a bias score representing the degree of bias and attribute labels indicating the cause of the bias as output. Specific examples of this output data include a bias score of 0.8 (high risk) relative to gender and a bias score of 0.2 (low risk) relative to the department of origin. This output is then sent to the subsequent alert notification department. In this subsequent processing, the system issues an immediate warning to the human resources department if the bias score exceeds a threshold. Through this process, the system mathematically detects unconscious human bias, forcibly maintaining the objectivity of evaluations. This system promotes organizational diversity and inclusion. This system prevents employee turnover due to inappropriate evaluations. As a specific application area, this system is used for auditing evaluation processes in global companies that propose promoting diversity. This system proactively prevents the risk of violating legal compliance. The generative AI analysis unit executes decision tree models for specific causes of bias in parallel. This generative AI analytics department automatically generates specific action plans to correct biases.

[0046] In the analysis of anonymous feedback, generative AI uses natural language processing techniques, such as keyword and sentiment analysis, to extract trust scores and improvement point lists for each manager, including terms like "trustworthy," "easy to negotiate with," "unfair," "insufficient support," "inadequate explanation," "overly harsh evaluation," and "lack of communication." For example, in cases with many negative feedback, the AI ​​summarizes the content and automatically generates specific improvement proposals such as "insufficient communication," "biased workload," "unclear evaluation criteria," and "imperfect support system." Furthermore, for example, by analyzing the temporal changes in feedback, if trust scores show a downward trend, the AI ​​determines that "early intervention is needed" and notifies managers and the HR department via an alert notification. Specifically, this generative AI analysis department executes an advanced natural language processing pipeline for analyzing unstructured text data. This generative AI analysis department obtains anonymous feedback text posted by employees as input data. Specific examples of this input data include strings of natural language such as "the supervisor's instructions are unclear, creating obstacles to business" and "always kindly accepting consultation." This generative AI analysis department uses a morphological analyzer and a sub-segmenter to segment the text data into a labeled sequence. This generative AI analysis unit inputs the labeled sequence into a pre-trained Transformer-based language model. This language model uses a self-attention mechanism to understand the context and computes semantic representation vectors for each label. The generative AI analysis unit uses these semantic representation vectors to classify sentiment polarity (positive, negative, neutral). The generative AI analysis unit further uses a named entity extraction algorithm to extract specific topics and keywords from the text. The generative AI analysis unit generates a trust score for each manager and a list of improvement points based on the extracted topics as output. As a specific example of this output data, the output includes a trust score of 0.45 and structured data such as "increase communication frequency" and "clear instructions" from the list of improvement points. The generative AI analysis unit forwards this output data to the subsequent evaluation report generation unit. In this subsequent processing, the system monitors the temporal changes in the trust score and initiates an alarm notification when a sharp decline is detected. Through this processing, the system converts a large amount of qualitative opinions into quantitative indicators. This system saves humans the trouble of reading all comments, instantly highlighting important issues. This system improves the organization's communication atmosphere and enhances psychological safety. In a specific application area, this system is used to monitor management quality in remotely-oriented organizations. Even in environments with limited face-to-face communication, the system accurately measures team well-being. This generative AI analytics department uses topic modeling to identify potential organizational issues. This generative AI analytics department automatically generates strategic recommendations for management.

[0047] The evaluation report generation department runs on a server and automatically generates evaluation reports for each manager, including turnover rate, transfer intention rate, promotion rate, bias detection results, trust score, and a list of improvement points obtained from the generative AI analysis department. For example, the report might record specific values ​​and comments such as "15% turnover rate in the past year," "5% transfer intention rate," "20% promotion rate for subordinates," "low-risk bias score," "0.8 trust score," "Improvement point: need to increase communication frequency," and "Anonymous feedback: 'Attentive support,' 'Strict evaluation,' 'Insufficient explanation.'" Reports are automatically updated periodically (e.g., monthly, quarterly) or in real-time and distributed from the server in a format accessible to the manager, their supervisor, and the HR department. Specifically, this evaluation report generation department performs data formatting processing to integrate various quantitative indicators and qualitative data output from the generative AI analysis department. This evaluation report generation department receives structured data in JSON format, including turnover rate, transfer intention rate, promotion rate, bias score, trust score, and a list of improvement points, as input data. This evaluation report generation department maps the structured data to a predefined report template. It executes a rendering engine to transform numerical data into visually understandable charts and graphs. For example, it plots the turnover rate as a line graph and bias scores as a radar chart. The department uses natural language generation algorithms to automatically generate explanatory text from the numerical data. It generates HTML or PDF format electronic documents as output. As a specific example of this output data, it generates dashboards with color-coded indicators and detailed report files for printing. This output document is then distributed to subsequent report viewing departments. In this subsequent processing, the system dynamically filters the displayed content based on user access permissions. Through this processing, the system provides complex analytical results in an intuitive and understandable form. This system significantly reduces the cognitive load required for data interpretation by managers and HR personnel. It supports rapid data-driven decision-making. As a specific application area, this system is used for the production of formal reports at quarterly HR evaluation meetings and management meetings. This system fully automates report production, dramatically improving the operational efficiency of the HR department. This evaluation report generation department has a built-in multilingual translation module. It provides reports translated into local languages ​​to various locations of companies worldwide in real time. The department tracks report viewing history and analyzes which metrics receive the most attention.

[0048] When the Alert Notification Department detects significant risks to the organization, such as a sharp increase in turnover and transfer intentions, a sharp drop in trust scores, or a high-risk bias score, it sends real-time alerts to management, HR, and operations. For example, alerts may include specific situations and recommended actions such as "Employee turnover has exceeded 20% in the past three months. Urgent action is needed" or "Most anonymous feedback received comments about 'unfair evaluations.' A reassessment of evaluation criteria is recommended." Specifically, this Alert Notification Department executes an event-driven background process that continuously monitors changes in the system's status. It receives various risk scores and indicator data streamed in real-time from the Generative AI Analysis Department as input data. Specific examples of this input data include an array of turnover prediction scores updated every second and daily bias score values. The Alert Notification Department queries this input data against an in-memory rule engine. This rule engine maintains multiple predefined threshold conditions. The Alert Notification Department determines whether the input data meets the threshold conditions through logical operations. If the conditions are met, the Alert Notification Department generates an alert message indicating specific actions as output. As a specific example of this output data, a text message stating "The risk of turnover in Manager D's team is increasing sharply" and importance indicators are generated. This alert notification department passes this output message to subsequent communication modules, sending it to target users as an email, chat tool, or push notification via a dedicated application. In this follow-up processing, the system tracks the alert's activation and response status, escalating notifications to higher management in cases of non-response. Through this process, the system detects potential crises within the organization before they materialize, enabling rapid intervention. This system prevents organizational collapse due to neglecting problems. This system reduces management costs through the automation of risk management. As specific application areas, this system is used for the early detection of harassment and relief for those with poor mental health. This system functions as a critical infrastructure for protecting employee safety and well-being. This alert notification department uses machine learning to learn from past alert response history. This alert notification department has adaptive threshold adjustment capabilities to reduce false detections and filter only truly important alerts. This alert notification department optimizes the timing and frequency of notifications to match users' behavior patterns.

[0049] The user interface on the client side allows managers, employees, HR managers, and other users to access the server via a browser or dedicated application to view evaluation reports, publish feedback, conduct evaluation interviews, and confirm improvement proposals. For example, managers can review their own evaluation reports, understand their trust scores and areas for improvement, and thus re-evaluate management actions. Employees can anonymously publish feedback and verify their past feedback history and career development. HR managers can understand the organization's overall evaluation biases and potential risks, using this information as evidence for system improvements. Specifically, this user interface is installed as a front-end application running on the client terminal. It renders dynamic web components built using HTML, CSS, and JavaScript. It establishes secure WebSocket or HTTPS communication with the server. It performs single sign-on processing using user authentication information. It receives user events such as mouse clicks, keyboard inputs, and touch operations as input data. Specific examples of this input data include changes to dropdown menus for specific report periods and keystrokes for inputting feedback text. This user interface converts these input events into asynchronous requests and sends them to the relevant processing units on the server side. This user interface receives response data from the server and uses a virtual DOM to rapidly redraw only specific portions of the screen. As a concrete example of this output data, a chart reflecting the latest evaluation score and newly arrived alert badges are displayed on the screen. In subsequent processing, the system logs the user's operation history and uses it to improve the interface. Through this process, the system provides users with a smooth, lag-free operating experience. Although this is a complex data analysis system, it achieves intuitive operability. This system improves user system utilization. Specifically, this system is suitable for a company-wide portal used by employees with varying IT skills. The system employs a responsive design, supporting access from various devices such as smartphones and tablets. This user interface adheres to accessibility standards, providing screen reader support for visually impaired users and a high-contrast mode. This user interface has a caching function that allows partial data viewing even offline.

[0050] The Feedback Input Department provides an interface for employees to anonymously post feedback to managers. For example, it includes multiple questions and free-response sections such as "Support System for Supervisors," "Fairness of Evaluations," and "Frequency of Communication," allowing employees to express frank opinions. Posted feedback is immediately sent to the server and analyzed by the Generative AI Analysis Department. Specifically, this Feedback Input Department implements a dedicated module for securely and reliably collecting qualitative text data from users. This Feedback Input Department receives text strings and selection-based responses entered by employees in an input form as input data. Specific examples of this input data include selection values ​​for radio buttons in a 5-stage evaluation and free-response text of up to 2000 characters. This Feedback Input Department performs input validation on the client side to prevent the infiltration of illegal characters and scripts. This Feedback Input Department uses public-key cryptography to encrypt the input data. This Feedback Input Department uses an asynchronous communication protocol to send the encrypted data to the server. This Feedback Input Department completely separates or hashes the user's identification information during transmission, mathematically guaranteeing anonymity. This Feedback Input Department generates a status code indicating completion of transmission as output and returns it to the User Interface Department. As a specific example of the output data, a JSON object with an HTTP status code of 200 and a success message is generated. In subsequent processing, the system decrypts the received encrypted data in a secure area and passes it to the generative AI analysis department for natural language processing. Through this processing, the system provides employees with a secure environment where they can express their opinions frankly without fear of repercussions. The system ensures high-quality raw data for extracting the organization's real issues. The system simultaneously satisfies data integrity and confidentiality. As specific application areas, this system is suitable for internal whistleblowing windows for compliance and violations, and direct suggestion boxes for management. The system enhances the organization's self-cleaning capabilities. This feedback input unit has an automatic backup function that temporarily saves data during input to local storage. This feedback input unit prevents data loss even when the network is down. This feedback input unit accepts multilingual input and supports automatic translation on the server side.

[0051] The Report Reading Department provides managers and HR personnel with an interface to view evaluation reports distributed from the server. For example, it displays information such as a chart showing the trend of turnover rate over the past year, a time series of changes in transfer intention rates, an overview of promotion performance, a heatmap of bias detection results, a chart showing the trend of trust scores, and details of the list of improvement points in a visually easy-to-understand format. Specifically, this Report Reading Department executes a data visualization engine that transforms complex structured data distributed from the server into a visually interpretable form. This Report Reading Department receives JSON-formatted report data as input data from the Evaluation Report Generation Department. Specific examples of this input data include a numerical array of turnover rates for the past 12 months and multidimensional matrix data representing bias detection results. This Report Reading Department parses the input data and calls upon the best charting library. For example, it generates line charts with interactive zoom functionality for time series data and heatmaps and scatter plots for multidimensional data. This Report Reading Department generates SVG or Canvas elements rendered in a browser as output. As a specific example of the output data, a dynamic chart component is generated that displays detailed values ​​when the mouse hovers over the data. In subsequent processing, when the user clicks on a specific data point on the chart, the system requests the detailed data behind it from the server (drill-down data). Through this processing, the system visually highlights patterns and outliers hidden within massive datasets. This system leverages the processing power of the human visual cortex, dramatically improving the speed of data understanding. This system enables even users with limited data literacy to utilize highly sophisticated analytical results. As a specific application area, this system is suitable for executive dashboards that allow management to grasp the overall soundness of the organization at a glance. This system always displays real-time updated information. The report viewing section provides the ability to customize dashboard layouts according to user preferences. The report viewing section has the ability to export specific charts as image files or PDFs. The report viewing section employs delayed read technology to minimize data retrieval time.

[0052] The Performance Appraisal and Interview Support Department assists managers and their subordinates in developing specific action plans during regular performance appraisal interviews, while simultaneously referencing evaluation reports, feedback records, and AI-based improvement proposals. For example, it provides prompts for AI-generated topics and action lists such as "Analysis of the Causes of Increased Turnover Rate This Period," "Background of Decreased Trust Score," and "Specific Countermeasures to Strengthen Career Support," thereby improving the quality of performance appraisal interviews. Specifically, this department implements a real-time synchronization module to structure communication between managers and employees and promote effective dialogue. The department receives past evaluation data and feedback records from the managers and employees being interviewed, as well as a list of improvement proposals generated by the generative AI analysis department, as input data. Specific examples of this input data include numerical data indicating the degree of achievement of goals set in the previous interview and the AI-extracted keyword "inadequate communication." Based on this input data, the department generates a dynamic agenda to guide the interview. The department also generates an interactive workspace screen shared on the interface as output. As a specific example of the output data, the system displays an input form for action plans that both parties can edit simultaneously, and a progress bar indicating the discussion's progress. In subsequent processing, the system saves the agreed-upon action plans to the database management department and reuses them as input data for the next interview. Through this process, the system standardizes the interview process, which can easily become subjective, and enforces a high-quality feedback loop. This system prevents interviews from becoming mere formalities and promotes substantive action-based change. The system uses objective data to eliminate cognitive biases between superiors and subordinates. As a specific application area, this system is used as a platform for regular 1-on-1 meetings in companies that have implemented goal management and OKR systems. In online interviews in remote work environments, the system provides an environment where participants can view the same data and discuss simultaneously through screen sharing. This assessment interview support department integrates with speech recognition AI, enabling real-time text-to-text conversion of interview speech to automatically generate meeting minutes. This assessment interview support department analyzes changes in the speaking ratio and emotional expression during the interview, providing real-time prompts for better communication.

[0053] Furthermore, for example, when managers implement improvement actions, the content is reported from the terminal to the server, and AI monitors subsequent changes in turnover rate and trust scores, enabling quantitative evaluation of the improvement effect. For instance, the effectiveness verification of improvement measures such as "increasing communication frequency to once a week resulted in a trust score increase from 0.6 to 0.8" can also be automated. Specifically, this system executes a feedback loop mechanism to track the implementation status of improvement actions and the resulting changes in organizational indicators. The system acquires implementation records of improvement actions input by managers from the terminal, as well as action data and feedback data from subordinate employees over a subsequent period, as input data. Specific examples of this input data include a text log stating "weekly 1-on-1 meetings were implemented," and a time-series array of trust scores collected over the following month. The system inputs this input data into a causal inference algorithm or reinforcement learning model. The model statistically infers the causal relationship between the implemented actions and changes in indicators. The system generates an effect quantity score representing the effectiveness of the improvement measures and an inferred return on investment value as output. As a specific example of the output data, the system calculates values ​​such as a 0.2-point increase in trust score based on countermeasures and a 15% reduction in turnover risk. This output result is fed back to the subsequent generative AI analysis department as learning data to improve the accuracy of future improvement proposals. Through this process, the system goes beyond simple data visualization, scientifically verifying which actions actually yield results. The system automatically extracts best practices from across the organization and accumulates them as a knowledge base. This system enables the high-speed operation of the PDCA cycle for personnel countermeasures. Specific application areas include company-wide organizational culture reform projects and the effectiveness measurement of leadership training for management positions. The system quantitatively demonstrates whether training recipients are correlated with changes in actual management actions. The system supports early termination of ineffective countermeasures and data-driven business judgment that focuses resources on effective countermeasures. The system has the capability to evaluate multiple improvement countermeasures in parallel using an A / B testing framework.

[0054] Thus, the implementation of this invention employs a structure where data collection, multi-angle analysis based on generative AI, evaluation report generation, alarm notification, and database management are uniformly executed on the server side, while users on the terminal side can view reports, publish feedback, conduct evaluation interviews, implement improvement actions, and report. This structure ensures fair and transparent personnel evaluation for managers, creating a system that contributes to improved trust and performance across the organization. Generative AI plays a core role in the system, including comprehensive data analysis, bias detection, feedback analysis based on natural language processing, prediction of career advancement potential, automatic generation of improvement proposals, and verification of the effectiveness of improvement measures. This eliminates the previous subjective and individualized evaluations, making it possible to achieve a personnel evaluation system that is both objective and transparent. Specifically, this system uses a distributed architecture to link various processing units, executing a fully automated integrated control module that streamlines the entire process from data collection to analysis, visualization, and action verification. The system continuously streams and receives diverse raw data from all digital touchpoints within the organization as input data. Specific examples of this input data include daily attendance records, email logs, chat message history, and periodic questionnaire responses. This system inputs this data into large-scale language models and deep learning models, which form the core of the generative AI analytics department, performing feature extraction and inference in high-dimensional space in real time. The system outputs comprehensive dashboard data representing organizational soundness, individual risk alerts, and optimized action plans. As a specific example of this output data, it generates updated results for a company-wide turnover prediction model and personalized management guidelines for individual managers. In subsequent processing, the system pushes and distributes this output data to users' terminals, facilitating organizational-wide action change. Through this processing, the system processes personnel data on a scale and speed exceeding the limits of human cognition, establishing an extremely objective evaluation basis that eliminates subjectivity and bias. This system dramatically improves the quality of evaluations while reducing the enormous time and cost spent on the personnel evaluation process. This system increases employee engagement, ultimately contributing directly to improved corporate performance. As a specific application area, this system is used in listed companies promoting human capital management as the basis for highly transparent information disclosure to investors. In the context of ESG investing, this system provides objective evidence to demonstrate a company's social sustainability. This system is built using cloud-native container technology and features automatic scaling capabilities to dynamically expand and shrink computing resources as needed. It also boasts a continuous integration environment capable of deploying the latest AI algorithms without downtime.

[0055] (System Composition) The system described in this implementation includes a data collection department, a database management department, a generative AI analysis department, an evaluation report generation department, an alarm notification department, a user interface department, a feedback input department, a report viewing department, and an evaluation interview support department. The data collection department automatically and periodically collects data from multiple internal and external systems and databases within the organization, such as the HR management system, business management system, project management system, employee questionnaire system, anonymous feedback system, attendance management system, company social networking site, email system, and training management system, through API linkage, batch processing, and file import. For example, it collects comprehensive information about employees' subordinates from multiple angles, including their departure date and transfer application date, whether they have engaged in job hopping, past promotion history, performance evaluation scores, potential scores, anonymous feedback comments, 360-degree evaluation results, attendance data such as lateness, early departure, overtime, paid leave status, project participation history, submission of business deliverables, number and content of posts on company social networking sites, number of emails sent and received, training history, qualification status, records of self-motivation activities, health diagnosis results, and stress test results. For example, it can obtain detailed data such as whether an employee has expressed a transfer request twice in the past year, attended three training sessions, served as a project leader once, and averaged 30 posts per month on company social media. Specifically, this system consists of multiple independent component groups based on a microservices architecture. This data collection unit executes a data ingestion pipeline running on the edge nodes responsible for interfaces with external systems. This data collection unit obtains REST API response payloads, batch streams of CSV files, and database transaction logs as input data. Specific examples of this input data include JSON objects containing basic employee information obtained from the HR system and daily work hours CSV files exported from the attendance system. This data collection unit performs schema validation, data cleaning, and format conversion on this input data. For example, this data collection unit converts date strings to UNIX timestamps and fills in missing values ​​with the mean or median. This data collection unit generates normalized data records as output, conforming to the system's internal standard format. Specific examples of this output data include Parquet format files with all attributes typed and message topics from a distributed messaging system. The data collection department asynchronously transmits the output data to the subsequent database management and generative AI analysis departments. Through this process, the system unifies and integrates information from diverse data sources, building a robust data foundation for subsequent advanced analytical processing. This system eliminates data silos, enabling cross-organizational data utilization. By automating data collection, the system completely eliminates human input errors and data gaps. Specifically, this system is suitable for integrating personnel data in corporate groups with mixed systems due to mergers and acquisitions.This system accepts different data formats and quickly applies a unified evaluation benchmark. The data collection department employs change data capture technology that tracks data change history to achieve real-time data synchronization. This data collection department also utilizes data compression algorithms designed to save network bandwidth.

[0056] The Database Management Department organizes and manages the diverse data acquired by the Data Collection Department according to time series, enabling retrieval, extraction, and summarization from multiple perspectives, such as individual managers, employees, departments, and projects. For example, it builds the data foundation for detailed analysis, such as the turnover rate of employees under specific managers over the past three years, annual comparisons of transfer intention rates in a department, the distribution of performance evaluations of members in specific projects, each employee's career path history, visualization of cross-departmental talent mobility, time series changes in anonymous feedback, and correlation analysis between training experience and promotion performance. For instance, it can also extract periods of high transfer intention rates in each department and compare them with the timing of organizational restructuring. Specifically, this Database Management Department executes a distributed columnar database engine capable of processing petabyte-level data. This department receives streams of normalized data records transmitted from the Data Collection Department as input data. Specific examples of this input data include employee action logs arriving at a rate of thousands per second and datasets of evaluation scores loaded in batches daily. The database management department partitions the input data across multiple dimensions, including timeline, organizational hierarchy, and individual ID, and distributes it across storage nodes. To achieve high-speed summary queries, the department applies compression techniques such as dictionary encoding and run-length encoding to each column. The database management department generates a result set of SQL queries from the user interface department and the generative AI analysis department as output. Specific examples of this output data include a summary table, in milliseconds, representing the trend of transfer intention rates over the past five years within a specific department, and statistical data representing the variance of evaluation scores for members of a specific project. In subsequent processing, the system caches this result set in memory to further improve the response speed for the same query. Through this processing, the system enables real-time execution of complex, multidimensional analyses of massive resume data. The system eliminates the waiting time incurred during data extraction, providing an analytical environment that does not hinder thinking. To improve data availability and fault tolerance, the system replicates data to multiple geographic regions. Specifically, this system is used as the backend of a company-wide talent management system in a global enterprise with tens of thousands of employees. This system provides the data foundation needed for organizational restructuring simulations and the formulation of future layoff plans. The database management department automatically applies data lifecycle management strategies, automatically migrating legacy data to inexpensive cold storage. The database management department also installs role-based access control to prevent unauthorized access to confidential data.

[0057] The Generative AI Analytics Department utilizes data accumulated in the Database Management Department for diverse AI analyses, such as combining large-scale language models, gradient boosting models, random forests, neural networks, natural language processing models, clustering models, and anomaly detection models. The Generative AI Analytics Department calculates the turnover rate and transfer intention rate of each manager's subordinates. For example, it summarizes the number of employees who left and the number who expressed transfer intentions in the past year, dividing by the total number of subordinates to obtain quantitative indicators. For instance, if a manager has 20 subordinates, with 3 leaving and 2 expressing transfer intentions, the turnover rate is 15% and the transfer intention rate is 10%. Furthermore, regarding subordinates' promotion performance, the department calculates the number and promotion rate of employees promoted in the past two years, analyzing the performance evaluation scores, project leadership experience, and self-help activity records of promoted employees to evaluate the manager's career support performance from multiple perspectives. For example, if 5 subordinates are promoted, with 3 having experience as new project leaders and 2 having obtained qualifications, the AI ​​determines that "active career support was provided." The generative AI analytics department uses data such as employees' performance evaluation scores, potential scores, business results, project leadership experience, self-motivation activity records, training records, posts on company social media, health diagnoses, and stress test results as input to predict the likelihood of promotion within the next six months. For example, employees with high performance evaluations, active self-motivation activities, and good stress test results are assessed by the AI ​​as having a "high likelihood of promotion," and this is reflected in the manager's career support performance evaluation. Regarding the detection of bias in managers' evaluations of subordinates, the AI ​​uses statistical and clustering methods. For instance, it detects bias and calculates bias scores when evaluation scores are extremely skewed or when there is a bias towards specific attributes (gender, age, job title, department of origin, etc.). For example, if a manager consistently gives low evaluations to female employees or high evaluations to employees from a specific department, the AI ​​assesses this as a "risk of bias" and records the specific evaluation bias in the report. In the analysis of anonymous feedback, generative AI uses natural language processing techniques, such as keyword and sentiment analysis, to extract trust scores and improvement point lists for each manager, identifying keywords like "trustworthy," "easy to negotiate with," "unfair," "insufficient support," "inadequate explanation," "overly harsh evaluation," and "lack of communication." For example, in cases with many negative feedback, the AI ​​summarizes the content and automatically generates specific improvement proposals such as "insufficient communication," "biased workload," "unclear evaluation criteria," and "imperfect support system." Furthermore, by analyzing the time-series changes in feedback, when trust scores show a downward trend, the AI ​​determines that "early intervention is needed" and notifies managers and the human resources department via an alert notification.Specific examples of prompts required for executing the generative AI analysis department include: "Calculate the turnover rate, transfer intention rate, and promotion rate of Manager A's subordinates over the past year, perform bias detection and sentiment analysis of anonymous feedback, and extract trust scores and improvement points," "Predict the likelihood of promotion in the next six months based on employee B's performance evaluation score, self-motivation activity history, and stress test results," "Analyze the evaluation data of Manager C's subordinates by each attribute, detect any evaluation bias, and calculate bias scores," and "Analyze the text data of anonymous feedback through natural language processing, summarize negative opinions, and generate improvement proposals," etc. Specifically, this generative AI analysis department executes an inference pipeline that orchestrates multiple machine learning models. This generative AI analysis department receives multidimensional tensors and natural language prompts extracted from the database management department as input data. Specific examples of this input data include a numerical matrix representing the employee's action history and an instruction string stating "Calculate the turnover rate of Manager A's subordinates over the past year and perform bias detection." This generative AI analysis department uses a word segmenter to convert this prompt text into a token sequence. This generative AI analysis unit inputs the labeled sequence and numerical tensor into a multimodal, large-scale language model. This large-scale language model, based on the Transformer architecture, uses a self-attention mechanism to compute contextual dependencies between labels and data. This large-scale language model has billions of parameters and is pre-trained and fine-tuned specifically for the human resources domain. The generative AI analysis unit generates structured text data and an array of predicted scores as output. As a specific example of this output data, a string of key-value pairs following JSON format is generated, containing specific values ​​such as a turnover rate of 15%, a bias score of 0.8, and the improvement proposal "eliminating communication deficiencies." The generative AI analysis unit parses this output data and passes it to the subsequent evaluation report generation and alert notification departments. In this subsequent processing, the system renders dynamic reports based on this output data and triggers alerts based on threshold determinations. Through this processing, the system seamlessly performs complex data analysis and natural language generation using a single model. The system instantly generates analysis reports that are natural and easy to understand, resembling human writing. This system replaces the role of senior data scientists, drastically reducing analysis costs. In its specific application area, this system is used in mid-sized enterprises with insufficient human resources departments, serving as a system that automatically provides analytical insights equivalent to those of professional HR consultants. This system utilizes prompting engineering techniques to flexibly handle diverse analytical tasks aligned with user intent. To illustrate the basis of the inferences, this generative AI analytics department outputs attention weights, representing which parts of the input data were focused upon, as visual data.

[0058] The Evaluation Report Generation Department automatically generates evaluation reports for each manager, including data such as turnover rate, transfer intention rate, promotion rate, bias detection results, trust score, and a list of improvement points obtained from the Generative AI Analysis Department. For example, the report might include specific figures and comments such as "15% turnover rate in the past year," "5% transfer intention rate," "20% promotion rate for subordinates," "low-risk bias score," "0.8 trust score," "improvement point: need to increase communication frequency," and "anonymous feedback: 'very supportive,' 'strict evaluation,' 'insufficient explanation.'" Reports are automatically updated periodically or in real-time and distributed from the server in a format accessible to the manager, their supervisor, and the HR department. Reports can be generated in various formats, such as monthly automatically generated reports, ad-hoc reports for specific events (e.g., large-scale turnover), individual employee career assessment reports, and cross-departmental comparative reports. Specifically, this Evaluation Report Generation Department executes a rendering pipeline that converts structured data output from the Generative AI Analysis Department into visual documents. This evaluation report generation department receives a JSON object as input data, containing turnover rate, transfer intention rate, promotion rate, bias detection results, trust score, and a list of improvement points. Specific examples of this input data include the current values ​​of each indicator, past trend data, and improvement proposal text described in natural language. This evaluation report generation department injects this input data into a dynamic template engine described using HTML5 and CSS3. This evaluation report generation department uses data binding technology to map numerical data to the properties of various chart components. This evaluation report generation department generates an interactive webpage viewable in a browser or a printable PDF document as output. Specific examples of this output data include a report containing a heatmap that color-codes each manager's strengths and weaknesses, and an animated line graph showing changes over time. This evaluation report generation department distributes the output document to subsequent report reading departments and displays it on the terminals of designated users. In this subsequent processing, the system records the report's viewing log, analyzes which sections are viewed the longest, and optimizes the composition of future reports. Through this process, the system transforms dry numerical data into intuitively understandable storyboards. This system eliminates information asymmetry and improves organizational transparency. It completely eliminates manual work required for report creation, enabling real-time information sharing. In a specific application area, this system is used to automatically generate publicly available human capital data for board meetings and shareholder meetings. The system instantly provides auditable, accurate, and consistent reports. This evaluation report generation department has the ability to dynamically adjust the granularity of the displayed information based on the user's position and permissions. For example, it can emphasize a company-wide summary to management and a specific action list to on-site managers.

[0059] The Alert Notification Department sends real-time alerts to managers, HR, and management when significant risks to the organization are detected, such as a sharp increase in turnover or transfer intentions, a sharp drop in trust scores, or a high-risk bias score. For example, alerts may include specific situations and recommended actions such as "Employee turnover has exceeded 20% in the past three months. Early intervention is needed," "Most anonymous feedback received comments about 'unfair evaluations.' A reassessment of evaluation criteria is recommended," "Bias score is high-risk. The evaluation method needs to be re-examined," and "Trust score is below 0.4. Enhanced communication is recommended." Specifically, this alert notification department employs an event-driven asynchronous processing architecture. It receives anomaly detection signals from various modules via a message queue system. The department evaluates the received signals as multi-dimensional vectors. It executes a threshold determination algorithm. It uses dynamic thresholds to reduce false detections. It calculates the priority of notifications. It dynamically switches communication protocols based on priority. It optimizes network bandwidth consumption. It ensures real-time performance. This alert notification unit works in conjunction with a push notification server. It generates a payload in a structured data format. It encrypts and sends the payload. It monitors the received acknowledgment responses. It executes a retransmission control algorithm. This alert notification unit improves communication reliability. It enhances the availability of the entire system. This alert notification unit provides the technical benefit of early detection of personnel risks. This alert notification unit is particularly effective in large-scale enterprise network environments.

[0060] The User Interface (UI) department provides access to the server via a browser or dedicated application for managers, employees, HR personnel, and other users to view evaluation reports, publish feedback, conduct evaluation interviews, and confirm improvement proposals. For example, managers can review their own evaluation reports, understand their trust scores and areas for improvement, and thus reassess management actions. Employees can anonymously publish feedback, verifying past feedback records and their career development. HR personnel can assess the organization's overall evaluation biases and potential risks, using this information as evidence for system improvements. For instance, it provides various interfaces such as dashboards displaying key metrics, detailed individual reports, feedback submission screens, and evaluation interview topic displays. Specifically, this UI department employs a single-page application architecture. It uses a virtual document object model to accelerate screen rendering. It communicates with the server via an application programming interface (API). It sends and receives objects in structured data formats. It utilizes asynchronous communication protocols. It hides network latency. It builds state management storage in client-side memory. It detects state changes and dynamically re-renders screen components. This user interface utilizes data binding technology. This user interface reduces response time to user actions. This user interface leverages hardware acceleration from the graphics processing unit. This user interface smoothly renders complex data visualization components. This user interface dynamically changes its layout according to the device's screen size. This user interface implements responsive design. This user interface reduces the cognitive load on users. This user interface is widely used in enterprise environments with multiple terminals.

[0061] The Feedback Input Department provides an interface for employees to anonymously post feedback to management. For example, it includes multiple questions and free-response sections such as "Regarding the Support System for Supervisors," "Regarding the Fairness of Evaluations," and "Regarding the Frequency of Communication," allowing employees to express frank opinions. Posted feedback is immediately sent to the server and analyzed by the Generative AI Analysis Department. For example, it incorporates designs that enhance employee psychological security, such as a combination of selective questions and free-response input screens, a confirmation screen for past feedback history, and automatic response message display after feedback is posted. Specifically, this Feedback Input Department implements a secure data transmission protocol. It encrypts the input text data on the client side. It uses public-key cryptography. It prevents man-in-the-middle attacks. It manages each field of the input form as an independent component. It counts the number of input characters in real time. It performs input validation on the client side. It prevents sending invalid requests to the server. It conserves network bandwidth. It serializes input data into a structured data format payload. This feedback input unit uses asynchronous communication to send the payload to the server. It eliminates local cached data after transmission. It prevents confidential information from remaining on the terminal. It technically guarantees anonymity. It uses a hash function to irreversibly transform the user identifier. It verifies data integrity. It provides a technical reduction in the risk of information leakage. This feedback input unit is particularly useful in financial institutions and medical institutions with stringent compliance requirements.

[0062] The Report Viewing Department provides managers and HR personnel with an interface to view evaluation reports distributed from the server. For example, it displays information such as turnover rate trends over the past year, time-series changes in transfer intention rates, a summary of promotion performance, heatmaps of bias detection results, trust score trends, and detailed lists of improvement points in a visually easy-to-understand format. It utilizes various display formats, including charts, heatmaps, text summaries, and comparison tables. Specifically, this Report Viewing Department employs a dedicated rendering engine for data visualization. It parses multidimensional data arrays received from the server. It converts the data arrays into vertex data for graphical rendering. It calls standard web-based graphical application programming interfaces (APIs). It parallelizes the rendering process using the client terminal's graphics processing unit. It renders scatter plots and heatmaps containing tens of thousands of data points in milliseconds. It detects user scrolling and zooming actions. It dynamically recalculates the data in the display area based on user actions. It executes a level-of-details control algorithm. This report viewing department adjusts the aggregation level of data based on the display magnification. This report viewing department keeps the rendering load constant. This report viewing department optimizes garbage collection to prevent memory leaks. This report viewing department ensures stable operation even during long viewing sessions. This report viewing department delivers the technical benefit of making large amounts of data readily apparent. This report viewing department is extremely effective in dashboard systems that provide management with a comprehensive overview of the organization's status.

[0063] The Performance Appraisal and Interview Support Department assists managers and subordinates in developing specific action plans while conducting regular performance appraisals, referring to evaluation reports, feedback records, and AI-based improvement proposals. For example, it provides AI-generated topics and action lists such as "Analysis of the Causes of Increased Turnover Rate This Period," "Background of Declining Trust Score," and "Specific Countermeasures to Strengthen Career Support," improving the quality of performance appraisals. Features include automatically listing topics before the interview, displaying past improvement actions and their effects in chronological order, recording notes and action plans in real-time during the interview, and automatically notifying follow-up tasks after the interview. Specifically, this Performance Appraisal and Interview Support Department implements a real-time synchronous communication protocol. It establishes a WebSocket connection. It synchronizes status across multiple client terminals in milliseconds. It executes an exclusive control algorithm. It prevents data conflicts caused by simultaneous editing. It records operation history as a time-series transaction log. It only transmits differential data over the network. This assessment and interview support department minimizes communication data volume. It executes a natural language processing model in the background. It analyzes the input note text in real time. It applies a keyword extraction algorithm. Based on the extracted keywords, it retrieves relevant past action plans from the database. It dynamically displays the search results as a recommendation list on the screen. It reduces the user's input load. At the end of the interview, it extracts all action items as structured data. It automatically registers them in the task management database. This assessment and interview support department delivers a seamless technological effect for business processes. It is particularly effective in online interviews in remote work environments.

[0064] Thus, the system described in this implementation achieves data collection, data management, AI analysis, report generation, alarm notification, user interface, feedback input, report viewing, and evaluation interview support through the coordinated operation of various components. It uses generative AI as its core to improve the fairness, transparency, and objectivity of personnel evaluation, thereby enhancing trust and performance across the entire organization. Specifically, this system operates on a distributed processing foundation based on a microservices architecture. Each component is deployed as an independent container. Resources are dynamically allocated using container orchestration tools. The number of instances of each module is automatically increased or decreased based on load conditions. Automatic scaling is implemented. The overall system processing capacity is optimized. Service mesh technology is incorporated into inter-module communication. Communication encryption and authentication are transparently performed. A zero-trust network security model is implemented. Single points of failure are eliminated. High availability is ensured. Inference processing of the generative AI model is offloaded to dedicated hardware accelerators. The load on the central processing unit is reduced. The overall system response speed is improved. This system delivers the technological benefits of automating and streamlining the data-driven personnel evaluation process through computer technology. This system is extremely useful in global enterprises with tens of thousands of employees, for use cases that simultaneously achieve standardized and efficient evaluation.

[0065] (Implementation steps) Step 1: Comprehensive collection of personnel-related data In this step, the data collection department automatically and regularly collects data on managers and their subordinates from various information sources both inside and outside the organization. Specifically, it obtains basic employee information, departure dates, transfer request dates, job-hopping activity, promotion history, performance evaluation scores, and potential scores from the HR management system; and project participation history and delivery status from the business management system. Furthermore, it collects data on lateness, early departure, overtime, and paid leave from the attendance management system; training participation history and qualification status from the training management system; and the number of messages and content, and the number of sent and received emails from the company's internal social networking and email systems. In addition, it obtains 360-degree feedback results, free-response feedback comments, health diagnosis results, and stress assessment results from anonymous feedback systems and employee questionnaires, accumulating multi-faceted and comprehensive information about managers' subordinates. For example, it can obtain detailed data such as whether an employee has requested a transfer twice in the past year, participated in three training sessions, served as a project leader once, and has an average of 30 messages per month on the company's internal social networking system. Specifically, this data collection unit executes a distributed crawler engine. It sends connection requests to multiple external systems in parallel. It dynamically generates request headers according to the application programming interfaces (APIs) of each system. It securely manages authentication tokens. It applies rate-limiting control algorithms. It prevents excessive load on external systems. It stores the acquired raw data in a temporary in-memory data grid. It performs data cleaning. It fills in missing values ​​and removes outliers. It applies a normalization algorithm to unify the data format. It removes control characters and Hypertext Markup Language (HTML) tags that constitute noise from unstructured text data. It converts the extracted data into documents in a structured data format. It assigns timestamps and unique identifiers to the converted documents. It publishes the documents to a message queue for transmitting data to subsequent processing modules. This data collection unit automates the pipeline processing from data collection to normalization. This data collection unit significantly reduces the computation time required for data preparation. This data collection department has become an indispensable foundational technology in the digital transformation of enterprises with a mix of legacy systems.

[0066] Step 2: Time-series storage and management in the database In this step, the Database Management Department organizes and manages the diverse data collected in the previous step according to time series, enabling retrieval, extraction, and summarization from multiple perspectives, such as individual managers, employees, departments, and projects. For example, it builds the data foundation for detailed analysis, including trends in employee turnover rates under specific managers over the past three years, annual comparisons of departmental transfer intention rates, distribution of performance evaluations for members in specific projects, career path profiles for each employee, visualization of cross-departmental talent mobility, time series changes in anonymous feedback, and correlation analysis between training experience and promotion performance. Specifically, the Database Management Department employs a distributed columnar database system. It implements a log structure merging tree to maximize write performance of time series data. The department temporarily stores received data in a memory buffer. When the buffer reaches a certain size, it refreshes the data as an immutable data file on disk. The department performs data file compression in the background. It optimizes read performance. The department physically configures the data as independent columns for each data attribute. This database management system reads only the necessary columns from disk when executing summary queries targeting specific attributes. It drastically reduces disk I / O bandwidth consumption. It optimizes each column and applies data compression algorithms. It uses run-length encoding and dictionary-based compression. It saves storage capacity. It builds multidimensional index structures. It uses space-filling curves to map multidimensional data to one-dimensional space. It processes complex range search queries at high speed. It delivers the technical advantage of reducing response time for analytical queries on large datasets to milliseconds. It is extremely effective as the backend for business dashboards that require real-time decision-making.

[0067] Step 3: Multi-faceted Analysis of HR Data via Generative AI In this step, the Generative AI Analysis Department performs various AI analyses using data accumulated in the Database Management Department. First, it calculates the turnover rate and transfer intention rate of each manager's subordinates. For example, it summarizes the number of employees who left and the number who expressed transfer intentions in the past year, dividing by the total number of subordinates to obtain quantitative indicators. Next, regarding subordinates' promotion performance, it calculates the number of employees promoted and the promotion rate in the past two years, analyzing the performance evaluation scores, project management experience, and self-motivation activity records of promoted employees to evaluate the manager's career support performance from multiple perspectives. Furthermore, using subordinates' performance evaluation scores, potential scores, business results, project management experience, self-motivation activity records, training experience, posts on company social media, health diagnoses, and stress assessment results as input, it predicts the likelihood of promotion within the next six months. Regarding the detection of bias in managers' evaluations of subordinates, AI uses statistical and clustering methods to detect bias and calculate bias scores when the distribution of evaluation scores is extremely skewed or when there is a bias towards specific attributes (gender, age, job title, department of origin, etc.). In the analysis of anonymous feedback, generative AI uses natural language processing techniques, such as keyword analysis (e.g., "trustworthy," "easily negotiable," "unfair," "insufficient support," "insufficient explanation," "overly harsh evaluation," "lack of communication"), to extract trust scores and a list of areas for improvement for each manager. Specific examples of prompts for generative AI to read include: "Calculate the turnover rate, transfer intention rate, and promotion rate of Manager A's subordinates over the past year; perform bias detection and sentiment analysis of anonymous feedback; and extract trust scores and improvement points." Other examples include: "Predict the likelihood of promotion within the next six months based on Employee B's performance evaluation score, self-motivation activity history, and stress assessment results." "Analyze the evaluation data of Manager C's subordinates by attribute, detect any evaluation bias, and calculate bias scores." "Analyze the text data of anonymous feedback through natural language processing, summarize negative opinions, and generate improvement proposals." Specifically, this generative AI analysis department implements an ensemble learning architecture that integrates multiple neural network models. This generative AI analysis department receives multi-dimensional numerical vectors containing employee attribute information, evaluation history, and action logs as input data. For example, the input vector is a 128-dimensional tensor containing age, number of consecutive years of service, past three evaluation scores, average monthly overtime hours, and number of posts on the company intranet. This generative AI analysis department normalizes this input tensor and performs preprocessing by filling in missing values ​​with the average. This generative AI analysis unit inputs preprocessed tensors into a deep learning model. This deep learning model is a multilayer perceptron consisting of multiple fully connected layers and dropout layers. The model applies nonlinear activation functions to each layer to extract the complex interactions of features in the high-dimensional space.This deep learning model generates a continuous score between 0 and 1 as the output, representing the probability of promotion within the next six months. For example, an output score of 0.85 indicates a very high probability of promotion. This generative AI analysis unit passes this output score to a threshold determination module; if the score is above 0.7, it assigns a label to the candidate for promotion. Furthermore, this generative AI analysis unit executes a large-scale language model based on the Transformer architecture for natural language processing tasks. The input to this large-scale language model is a sequence of anonymized feedback text data, appended with position-encoded word embedding vectors. For example, the input data is text with ambiguous evaluation criteria and reduced positivity, converted into a 512-dimensional vector sequence. This large-scale language model uses a self-attention mechanism to compute contextual dependencies. This large-scale language model generates structured data as output, including sentiment polarity scores and extracted improvement keywords. For example, the output data includes a negative score of 0.9 and the keyword "clarification of evaluation criteria." This generative AI analysis unit forwards this output data to the evaluation report generation unit. This generative AI analysis unit periodically relearns the model's weight parameters using backpropagation. This generative AI analytics unit executes optimization algorithms to minimize the cross-entropy error function. It delivers the technological capability to achieve pattern recognition of high-dimensional data that surpasses human cognitive abilities. This generative AI analytics unit plays a central role in use cases that eliminate subjective evaluations and achieve data-driven, objective personnel evaluations.

[0068] Step 4: Automatic Generation and Distribution of Evaluation Reports In this step, the evaluation report generation department automatically generates an evaluation report for each manager, including data such as turnover rate, transfer intention rate, promotion rate, bias detection results, trust score, and a list of improvement points obtained from the generative AI analysis department. For example, the report might include specific values ​​and comments such as "15% turnover rate in the past year," "5% transfer intention rate," "20% promotion rate for subordinates," "low-risk bias score," "0.8 trust score," "Improvement point: need to increase communication frequency," and "Anonymous feedback: 'Attentive support,' 'Strict evaluation,' 'Insufficient explanation.'" The report is automatically updated periodically or in real-time and distributed from the server in a format accessible to the manager, their supervisor, and the HR department. For example, reports can be generated in various formats, such as monthly automatically generated reports, ad-hoc reports for specific events (e.g., large-scale turnover), individual employee career assessment reports, and cross-departmental comparative reports. Specifically, this evaluation report generation department implements a dynamic document generation algorithm that integrates a template engine and a data binding mechanism. This evaluation report generation unit receives analysis results data in structured data form from the generative AI analysis unit. It maps this data to a virtual object tree in memory. The unit reads a predefined report template from memory. It replaces placeholders in the template with the values ​​of the corresponding data nodes in the virtual object tree. It executes conditional branching logic. If a bias score exceeds a specific threshold, it dynamically inserts a special visual component for warning purposes into the report. It calls a charting library for visualizing numerical data. Using time-series data as input, it generates vector graphics in memory. It embeds the generated image data into the report document. It serializes the final document into Portable Document Format (PDO) or Hypertext Markup Language (HTML). It saves the generated file to a distributed file system. It records a Uniform Resource Locator (URL) indicating the file's location in a database. It sets an access control list. It uses encryption to prevent access from unauthorized users. This evaluation report generation department delivers the technological benefit of dramatically reducing the computing resources and time required for report production. It is particularly effective in large-scale enterprise environments where personalized feedback is simultaneously provided to thousands of managers.

[0069] Step 5: Early Risk Detection and Response via Alert Notifications. In this step, the Alert Notification Department sends real-time alerts to management, HR, and operations when it detects significant risks to the organization, such as a sharp increase in turnover or transfer intention rates, a sharp drop in trust scores, or a high-risk bias score. For example, alerts may include specific situations and recommended actions such as "Employee turnover has exceeded 20% in the past three months. Urgent action is needed," "Most comments about 'unfair evaluation' have been received in anonymous feedback. A reassessment of the evaluation criteria is recommended," "Bias score is high-risk. The evaluation method needs to be re-examined," and "Trust score is below 0.4. Strengthened communication is recommended." Specifically, this Alert Notification Department is equipped with a composite event processing engine. It monitors the continuous data flow from the Database Management Department and the Generative AI Analysis Department. It applies a sliding window algorithm. It maintains recent data points within a certain period in memory. It performs statistical anomaly detection algorithms on the data within the window. It calculates moving averages and standard deviations in real time. This alert notification department raises an anomaly flag when the current value deviates from the statistical normal range. This alert notification department evaluates a complex set of rules that combine multiple anomaly flags using logical operators. This alert notification department generates an alert object if the rules are met. This alert notification department assigns the alert object an importance level, the time of occurrence, an identifier for the relevant manager, and textual data recommending action. This alert notification department sends the generated alert object to the message broker. This alert notification department utilizes the message broker's publish-subscribe model. This alert notification department asynchronously distributes push notifications to the appropriate recipients' terminals. This alert notification department implements a retransmission control mechanism to ensure the reliability of distribution. If no confirmation of receipt is received within a certain time, this alert notification department switches to an alternative communication channel and retransmits the alert. This alert notification department brings the technical effect of detecting potential risks to the organization before they become apparent, enabling rapid response. This alert notification department functions as an early warning system to minimize the risk of compliance violations and mass turnover.

[0070] Step 6: Visualizing and Utilizing Information Through the User Interface In this step, the user interface provides an interface for managers, employees, HR managers, and other users to access the server via a browser or dedicated application to view evaluation reports, publish feedback, conduct evaluation interviews, and confirm improvement proposals. For example, managers can review their own evaluation reports, understand their trust scores and areas for improvement, and reassess management actions. Employees can anonymously publish feedback, review past feedback records, and track their career growth. HR managers can understand the organization's overall evaluation biases and potential biases, using this as evidence for system improvements. Specifically, this user interface is installed as a single-page application running on a web browser on the client terminal. This user interface uses Virtual Document Object Model (VDOM) technology to optimize screen redrawing. This user interface receives structured data from the server via an asynchronous communication protocol. This user interface stores the received data in a state management container on the client side. This user interface detects state changes and only performs differential updates to the relevant screen components. This user interface eliminates the need to reread the entire screen. This user interface suppresses the response latency to user actions to milliseconds. This user interface (UI) optimizes hardware acceleration for the graphics processing unit. It smoothly renders complex animations and data visualizations. It adheres to responsive web design principles. It dynamically acquires the terminal's screen resolution and aspect ratio. Based on these parameters, it recalculates the layout grid. It provides optimal display across various devices, including smartphones, tablets, and desktops. It reduces cognitive load and improves information comprehension. This UI plays a crucial role in the daily use of company portal systems by employees with varying levels of IT literacy.

[0071] Step 7: Feedback Submission and Evaluation Interview Implementation. In this step, the Feedback Input Department accepts anonymous feedback submissions from employees. The Evaluation Interview Support Department supports managers and subordinates in conducting regular evaluation interviews, helping them develop specific action plans while referring to evaluation reports, feedback history, and AI-based improvement proposals. For example, it provides functions such as automatically listing topics before the interview, displaying past improvement actions and their effects in chronological order, recording notes and action plans in real-time during the interview, and automatically notifying follow-up tasks after the interview. Specifically, the Feedback Input Department performs real-time cleanup on the client side for the input text data. It escapes special characters to prevent cross-site scripting attacks. It encrypts data using public-key cryptography before sending it to the server. It prevents eavesdropping on the communication path. The Evaluation Interview Support Department runs a WebSocket server on the server side. It establishes a two-way real-time communication channel between the terminals of managers and employees. It installs an operation conversion algorithm. This assessment interview support system automatically resolves data conflicts when multiple users are editing simultaneously. It inputs the text data recorded during the interview into a natural language processing model in the background. This model then automatically extracts candidate action items from the text. The extracted candidates are pushed as structured data to the client terminal. They are then dynamically displayed as a recommendation list on the client terminal screen. The action plan determined at the end of the interview is inserted into a task management table in a relational database. Reminders are scheduled based on specified dates. This assessment interview support system digitizes the interview process and improves follow-up accuracy. It is best suited for use cases in companies implementing management by objectives (MBO) systems to improve the credibility and effectiveness of evaluations.

[0072] (specific use case) For example, imagine a technical department within an organization where the department head conducts personnel evaluations of their subordinates. In this scenario, the data collection department automatically retrieves employee departure dates, transfer request dates, job-hopping activity, past promotion records, performance evaluation scores, and potential scores from the technical department's personnel management system. Furthermore, it collects project participation records, supervisor experience, and deliverable submission status from the project management system, and attendance data such as lateness, early departure, overtime, and paid leave status from the attendance management system. Additionally, it collects training attendance records and qualification status from the training management system, the number and content of messages from the company's social networking site, the number of emails sent and received from the email system, and feedback comments and 360-degree feedback results freely recorded by employees from the anonymous feedback system. This data is organized chronologically by the database management department and managed in a way that allows for retrieval, extraction, and summarization by department head, employee, and project. Specifically, this data collection department implements a dedicated connector set to handle the diverse single data sources unique to the technical department. This data collection department also calls the application programming interface of the source code management system. This data collection department acquires development activity data such as the number of submissions, pull requests created, and code reviews performed by each employee in structured data format. This department parses the log files of the continuous integration system. This department extracts build success rates and test coverage shifts as numerical data. This department obtains vulnerability resolution times and severity levels of responsible faults from the fault management system. This department reads this unstructured and semi-structured data into data frames in memory. This department performs conversion processing to unify data types. This department normalizes time zone differences to Coordinated Universal Time (UTC). This department combines multiple data frames using the employee's unique identifier as the key. This department converts the combined data into a multi-dimensional time-series tensor. This department batch inserts this tensor into a distributed columnar database. This department asynchronously rebuilds the database indexes. This department optimizes the execution speed of subsequent retrieval queries. This data collection department brings the technological effect of automatically building a data foundation for quantifying engineers' productivity and contributions. In software development organizations that adopt agile development, this data collection department is indispensable for achieving objective performance evaluation.

[0073] The generative AI analytics department uses this data to calculate the turnover rate, transfer intention rate, and promotion rate of each department head's subordinates. For example, it outputs quantitative indicators such as, "Of Department Head A's 20 subordinates, 3 left in the past year, and 2 expressed a desire to transfer, resulting in a turnover rate of 15% and a transfer intention rate of 10%." Furthermore, it uses subordinates' performance evaluation scores, potential scores, project leadership experience, self-motivation activity records, training records, posts on company social media, health diagnoses, and stress test results as input to predict their likelihood of promotion within the next six months, classifying "Employee C, with high performance evaluations, active self-motivation activities, and good stress test results, as having a high probability of promotion." Regarding the detection of bias in department heads' evaluations of subordinates, the AI ​​uses statistical and clustering methods, such as detecting bias risks like "Department Head A consistently gives low evaluations to female employees" or "gives high evaluations to those from specific departments," and calculates bias scores. In the analysis of anonymous feedback, generative AI uses natural language processing techniques to perform keyword and sentiment analysis on terms such as "trustworthy," "negotiable," "unfair," "insufficient support," "inadequate explanation," "overly harsh evaluation," and "lack of communication," extracting trust scores and a list of improvement points for each department head. Specifically, this generative AI analysis department executes a multimodal neural network optimized for the data characteristics of the technical department. This generative AI analysis department receives composite data as input, including employee resumes, code review text, and performance evaluation scores. For example, the input data consists of a 64-dimensional vector of numerical data and a sequence of 512-dimensional word embedding vectors representing the tokenized comments from code reviews. This generative AI analysis department processes this input data using fully connected layers and a Transformer encoder, respectively. The generative AI analysis department integrates the output feature maps of both in a combination layer. The generative AI analysis department inputs the integrated feature vector into a classifier with multiple hidden layers. This classifier generates a score representing the probability of promotion within the next six months and a probability distribution representing the risk of leaving the company as output. For example, the output score is 0.92, with a probability distribution of 80% low risk and 20% high risk. This generative AI analysis department passes this output score to the threshold determination module; if the promotion probability is above 0.8, a promotion candidate flag is raised. Furthermore, to detect evaluation bias caused by department heads, this generative AI analysis department executes an unsupervised learning-based density-based spatial clustering algorithm. This department plots the evaluation data and employee attributes in a high-dimensional space, detecting clusters with statistically significant low evaluations of specific attributes. Based on this detection result, the department calculates a bias score. The department uses a natural language processing model to analyze anonymous feedback, outputting sentiment polarity scores and topic classification labels. This department organizes these outputs into structured data and forwards them to the evaluation report generation department.This generative AI analytics suite delivers the technological advantage of discovering complex relationships that human intuition cannot fully capture through computation in high-dimensional spaces. It is particularly effective in use cases where evaluation is highly specialized and easily humanized, and where fair and transparent evaluation criteria are established.

[0074] Based on these analysis results, the evaluation report generation department automatically generates evaluation reports containing specific values ​​and comments such as "Department Head A's turnover rate over the past year was 15%, transfer intention rate was 10%, subordinate promotion rate was 20%, bias score was high-risk, trust score was 0.6, improvement points: need to increase communication frequency, anonymous feedback: 'attentive support,' 'strict evaluation,' 'insufficient explanation,'" and distributes them regularly or in real time to the department head, superiors, and the human resources department. The alert notification department sends real-time alerts to the department head, human resources department, and management when it detects significant risks such as "Department Head A's subordinate employee turnover rate exceeded 20% in the past three months," "most 'unfair evaluation' comments were received in anonymous feedback," "bias score was high-risk," or "trust score was below 0.4." Specifically, this evaluation report generation department receives structured data output from the generative AI analysis department as input. This evaluation report generation department maps this data into a virtual document object model in memory. This evaluation report generation department reads a dedicated template for the technology department from memory. The report generation department dynamically replaces variable placeholders within the template using mapped data. It executes a data visualization library. In memory, it renders a radar chart representing the distribution of bias scores and a line graph representing the shift in turnover rate as vector graphics. The rendered image data is then inlined into the document. The completed document is converted to a portable document format, encrypted, and saved to a distributed file system. The alert notification department monitors the metadata of the generated reports. It detects bias scores exceeding a pre-set threshold. Triggered by this detection, it generates an emergency alert message. Push notifications are asynchronously sent to department heads and HR terminals via a message queue system. The notification log is logged in a database, and confirmation of receipt is tracked. This alert notification department minimizes the delay from information generation to delivery. It supports proactive management to prevent the loss of skilled personnel in technical departments.

[0075] The User Interface Department enables department heads to review their performance evaluation reports, understand trust scores and areas for improvement, and reassess management actions. Employees can anonymously submit feedback, verifying past feedback records and their career development. HR managers can assess organizational evaluation biases and potential bias risks, using this information as evidence for system improvements. The Feedback Input Department allows employees to submit frank opinions using multiple questions such as "Regarding the support system for superiors," "Regarding the fairness of evaluations," and "Regarding the frequency of communication," as well as a free-response section. Submitted feedback is immediately sent to the server and analyzed by the Generative AI Analytics Department. The Report Viewing Department provides department heads and HR managers with an interface to view performance evaluation reports distributed from the server. It displays, in a visually appealing format, graphs of turnover rates over the past year, time-series changes in transfer intention rates, a summary of promotion performance, heatmaps of bias detection results, graphs of trust scores, and details of the list of areas for improvement. The Performance Evaluation Interview Support Department supports department heads and subordinates in conducting regular performance evaluation interviews, allowing them to develop concrete action plans while referring to evaluation reports, feedback records, and AI-based improvement proposals. Specifically, this user interface unit uses WebAssembly technology to perform advanced data processing on the client terminal. This user interface unit caches large-scale evaluation data received from the server into the client-side memory. For user filtering operations, this user interface unit recalculates data locally without communicating with the server. This user interface unit redraws the screen in milliseconds. This feedback input unit performs real-time input validation algorithms on each field of the input form. This feedback input unit prevents the intrusion of illegal character codes and script tags on the client side. This feedback input unit serializes the input data into a structured data payload and securely sends it to the server using transport layer security protocols. This report viewing unit utilizes the client terminal's graphics processing unit to dynamically render heatmaps of bias detection results. This report viewing unit immediately displays detailed tooltips based on the user's hover actions. This assessment interview support unit installs a real-time network communication protocol. This assessment interview support unit encrypts and transmits audio and video streams point-to-point between the department head and employee terminals. This assessment interview support unit transcribes the audio during the interview into text in the background. This assessment of interview support utilizes a natural language processing model to automatically extract action items from text and reflect them in real-time in a shared editor on the screen. This assessment of interview support delivers the technological benefit of dramatically reducing user workload and improving communication quality. In remotely-oriented technology departments, this assessment of interview support is best suited for implementing effective online meeting use cases.

[0076] Specific examples of prompts that generative AI can read include: "Calculate the turnover rate, transfer intention rate, and promotion rate of Department Head A's subordinates over the past year, perform bias detection and sentiment analysis of anonymous feedback, and extract trust scores and improvement points," "Predict the likelihood of promotion in the next six months based on employee B's performance evaluation score, self-motivation activity history, and stress test results," "Analyze the evaluation data of Department Head C's subordinates by each attribute, detect any evaluation bias, and calculate bias scores," and "Analyze the text data of anonymous feedback through natural language processing, summarize negative opinions, and generate improvement proposals," etc. Specifically, this system installs a prompt engineering module. This prompt engineering module dynamically embeds structured data extracted from the database into natural language prompt templates. This prompt engineering module optimizes text length by utilizing the input context window limitation of a large-scale language model. This prompt engineering module executes a tokenization algorithm. This prompt engineering module converts the input text into an array of sub-word tags. For example, the input data is a 1024-dimensional tag array containing the text "Calculate the turnover rate of Department Head A's subordinates over the past year and extract improvement points." This prompting module sends the tag array to the inference engine of a large-scale language model. This inference engine utilizes the tensor core of the graphics processing unit to highly parallelize matrix multiplication operations. This inference engine also accelerates the computation of the self-attention mechanism. As output, this inference engine generates structured text containing the requested analysis results. For example, the output data is structured text with turnover rate, bias score, and improvement points as keys, and strings like 0.15, 0.8, and increased feedback frequency as values. This prompting module parses the output text and converts it into a data object that can be processed by the subsequent evaluation report generation department. This prompting module executes error handling logic. This prompting module automatically regenerates prompts and retryes if the output format is incorrect. This prompting module brings the technical effect of effectively integrating the output of generative artificial intelligence into deterministic system processes. This prompting module automates complex analysis tasks, becoming a fundamental technology for improving the accuracy of personnel evaluations.

[0077] Thus, by using the system of this invention, personnel evaluations for department heads can move beyond the previous subjective and individualized assessments. It objectively and quantitatively visualizes turnover rates, transfer intention rates, promotion rates, bias detection, trust scores, and areas for improvement. Through evaluation reports, alert notifications, and assessment interview support, it significantly improves the fairness, transparency, and trustworthiness of the entire organization. Specifically, this system implements a pipelined processing architecture that integrates the above modules. The system uses event-driven automatic control to manage the data flow from data collection to analysis, report generation, and notification. The system asynchronously transfers data between processing steps via in-memory message queues. The system minimizes disk I / O and maximizes overall system throughput. The system temporarily stores the inference results of machine learning models in cache memory. The system eliminates recalculation of the same input, shortening response time. The system continuously transmits system logs to a centralized log management server. The system monitors system operation status and error rate in real time. The system executes automatic recovery scripts upon anomaly detection. The system minimizes system downtime. This system transforms traditional HR processes, which rely on subjective human judgment, into a data-driven and algorithm-based objective process. It mathematically reduces evaluation bias and optimizes overall organizational performance. In modern corporate environments with employees possessing diverse values, this system proves highly useful in providing a fair and transparent basis for HR evaluations.

[0078] (Application Example 1) The specific processing flow corresponding to Use Case 1 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. Furthermore, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0079] (Implementation methods for carrying out the present invention) The embodiments of the present invention will be described in more detail using a nursing site manager evaluation system as an example. This system consists of a server and a terminal. The server side includes a data collection unit, a generative AI analysis unit, an evaluation report generation unit, a database management unit, and an alarm notification unit. The terminal side includes a user interface unit, a feedback input unit, a report viewing unit, and an evaluation interview support unit.

[0080] The Data Collection Department automatically and regularly collects data on managers and their subordinates from various information sources both inside and outside the nursing facility. For example, it obtains data from the nursing staff's personnel management system, including hiring date, termination date, transfer intentions, job transfer history, training history, qualification status, years of continuous service, past performance evaluations, promotion history, and reward / punishment history. It also obtains data from the nursing record system, including nursing plan achievement rate, daily nursing records, nursing content for each user, frequency of nursing implementation, details of nursing content (e.g., feeding assistance, bathing assistance, excretion assistance, rehabilitation support, recreational implementation status), user satisfaction questionnaire results, each user's nursing satisfaction score, and records of direct gratitude and complaints from users. Finally, it obtains data from the incident / near-incident reporting system, including the number of incidents, date and time of occurrence, location, cause, response measures, recurrence prevention measures, involved employees, incident severity classification (minor, moderate, major), and follow-up status after the incident. As a record of communication among employees, we collect data on business contact systems and chat logs, meeting attendance records, business handover records, employee consultation / suggestion records, meeting minutes of regular meetings, and exchanges of thank-you / appreciation messages among employees. As a result of regular stress checks and health diagnoses, we also obtain stress levels, health risk scores, guidance from doctors, implementation status of health maintenance activities, and mental health counseling records. As a result of anonymous feedback and 360-degree evaluations from users or their families, we collect opinions expressed through freehand writing, scores from 5-stage or 10-stage evaluations, individual comments on specific employees, letters or emails from users' families, thank-you notes or complaints, feedback submission frequency, and changes in the time series of feedback from questionnaires and dedicated feedback systems. For example, we also obtain users' freehand writings such as "slow response at night," "very detailed explanations," "friendly staff," "very careful feeding assistance," and "feeling uneasy during bowel movement assistance," satisfaction scores from 5-stage evaluations, and specific requests from families such as "thank you for the staff's response" and "hope for more communication." Specifically, this data collection department operates in a distributed computing environment. This data collection unit executes on a server comprised of multiple computing nodes. It establishes connections with external systems using an asynchronous communication protocol. It receives encrypted data packets via a secure communication path. The received data packets are temporarily stored in a buffer area in memory. A syntax analysis algorithm is used to remove unwanted noise from the data. The extracted text data is then input into a natural language processing model. This natural language processing model employs a multi-layer Transformer architecture with a self-attention mechanism. The input data for this model is a two-dimensional tensor composed of word embedding vectors. This input data contains multiple 512-dimensional floating-point vectors. For example, the first input data contains a numerical vector representing the hiring date.The second input data contains vectors of transformed nursing record text. The third input data contains vectors of transformed event report text. The output data from this model is a high-dimensional vector from which feature quantities have been extracted. This output data is a numerical array that quantifies the semantic features of each text. For example, the first output data contains a value of 0.85 representing positive sentiment. The second output data contains a value of -0.42 representing negative sentiment. The third output data contains a value of 0.12 representing neutral sentiment. This data collection unit transmits the output feature vectors to the subsequent database management unit. This data collection unit sorts and saves the feature vectors in chronological order. Through this structure, this system automatically extracts useful features from unstructured data. This system eliminates the burden of data input caused by manual human intervention, significantly improving processing speed. This technology can be applied as a data integration foundation for large-scale nursing operators managing multiple facilities.

[0081] The Database Management Department organizes and manages this diverse data according to time series, enabling retrieval, extraction, and aggregation from multiple perspectives, including individual managers, employees, users, departments, facilities, and events. For example, it builds a data foundation for detailed analysis, such as the turnover rate of a specific manager's subordinates over the past three years, annual comparisons of transfer intention rates for a particular facility, time-series changes in the care quality score of a specific employee, monthly trends in the number of incidents, time-series changes in anonymous feedback, correlation analysis between training experience and care quality score, the relationship between employee communication frequency and incident rate, the correlation between user satisfaction and care plan achievement rate, the relationship between employee health risk score and turnover rate, and the relationship between the frequency of feedback from family members and the number of complaints. For instance, it can also extract periods of high transfer intention rates for each facility and compare them with the timing of organizational restructuring. Furthermore, extracting each employee's care content preferences and areas of expertise can be used for appropriate talent allocation and training program development. Specifically, the Database Management Department implements high-availability data storage built on a distributed file system. This database management department constructs an index structure specifically for time-series data. It utilizes parallel processing processors for high-speed processing of multidimensional retrieval queries. The department expands the input feature vectors into memory. It uses a clustering algorithm to group the data. This clustering model calculates the Euclidean distance between data points in a high-dimensional space. The input data for this model is a multidimensional tensor containing employee attribute information and care quality scores. This input data is a 256-dimensional floating-point vector. Specifically, the first input data contains a time-series array representing the shift in turnover rate. The second input data contains a numerical array representing the number of events. The third input data contains a numerical array representing the sentiment scores of anonymous feedback. The output data from this model consists of integer values ​​representing the identifiers of the clusters to which each data point belongs. This output data represents the grouping results based on data similarity. Specifically, the first output data contains identifier one representing a high-risk group. The second output data contains identifier two representing a low-risk group. The third output data contains identifier three representing a group to be observed. This database management department assigns the output cluster identifiers as metadata to each record. This database management department uses clustering identifiers to optimize subsequent retrieval processing. Through this structure, the system can instantly extract specific patterns from massive time-series data. This system makes it possible to discover complex relationships that were previously difficult to find in purely relational databases. This technology can be applied to comprehensive human resources systems aimed at employee performance analysis and risk management.

[0082] The Generative AI Analytics Department utilizes data accumulated in the Database Management Department for diverse AI analyses. As generative AI, it combines large-scale language models, gradient boosting models, random forests, neural networks, natural language processing models, clustering models, and anomaly detection models. First, it calculates the turnover rate and transfer intention rate of each manager's subordinates. For example, it summarizes the number of employees who left and the number who expressed a transfer intention in the past year, dividing by the total number of subordinates to obtain quantitative indicators. For instance, if a manager has 10 subordinates, with 2 leaving and 1 expressing a transfer intention, the turnover rate is 20% and the transfer intention rate is 10%. Regarding nursing quality, it comprehensively evaluates nursing plan achievement rate, user satisfaction questionnaires, training records, qualification status, diversity of nursing content, frequency of nursing implementation, each user's nursing satisfaction score, and records of user thanks / complaints, etc., outputting multiple indicators such as "the average nursing quality score of subordinates is 4.3," "the nursing plan achievement rate is 95%," "the average user satisfaction score is 4.5," and "the diversity score of nursing content is 0.8." Regarding incidents / near misses, the number of reports per employee is calculated, resulting in detailed metrics such as "incident rate of 0.2 incidents / person / year," "major incident rate of 0.05 incidents / person / year," and "incident recurrence rate of 10%." Specifically, this generative AI analysis department implements an ensemble learning architecture integrating multiple deep learning models. This generative AI analysis department runs gradient boosting trees and multilayer perceptrons in parallel. This generative AI analysis department reads multidimensional tensors extracted from the database management department into memory. This generative AI analysis department performs preprocessing to fill in missing values ​​with the mean. This generative AI analysis department uses a normalization algorithm to unify the scale of the data. This neural network uses nonlinear activation functions in the hidden layers to learn complex features. The input data for this model is a one-dimensional array containing employee action history and evaluation metrics. The input data is a 64-dimensional floating-point vector. As a specific example, the first input data contains the value 2 representing the number of departing employees. The second input data contains the value 0.95 representing the care plan achievement rate. The third input data contains the value 0.2 representing the incident rate. The output data from this model are floating-point numbers representing predicted values ​​for future turnover probability and quality of care. This output data is a probability score normalized to the range of zero to one. For example, the first output data contains a value of 0.85 representing the turnover probability. The second output data contains a value of 0.72 representing the likelihood of improved quality of care. The third output data contains a value of 0.15 representing the probability of a major event occurring. This generative AI analysis unit compares the output predicted scores with a predetermined threshold. If the predicted score exceeds the threshold, the generative AI analysis unit activates a subsequent alarm generation module. Through this structure, this system goes beyond simply summarizing past data and proactively predicts future risks.This system improves prediction accuracy by detecting minute signs imperceptible to humans through computation in high-dimensional space. This technology can be applied to measures to improve staff retention rates in medical settings and nursing facilities.

[0083] Furthermore, regarding anonymous feedback from users or their families, generative AI uses natural language processing to analyze the text, extracting keywords and emotions such as "trustworthy," "friendly," "detailed explanations," "slow response," "feeling uneasy," "dissatisfaction with nighttime responses," "employee smiles provide reassurance," "detailed feeding assistance," and "lack of care during toileting assistance." It then generates trust scores and improvement point lists for each manager. For example, in cases with many negative comments, it automatically generates specific improvement proposals such as "insufficient explanations from employees," "anxiety about nighttime responses," "few conversations during feeding assistance," and "insufficient care during toileting assistance." Additionally, it analyzes the distribution of employee evaluation data and the tendencies of each attribute, such as detecting bias risks like "consistently giving low ratings to female employees," "giving high ratings to specific qualification holders," "strict ratings to newly hired employees," and "lenient ratings to long-term continuous workers," and calculates bias scores. AI also analyzes the relationship between employee health risk scores and stress test results with turnover and incident rates, extracting causal relationships such as "employees with high stress scores have a 1.5 times higher turnover rate than average" and "employees with high health risks have a 2 times higher incident rate." Specifically, this generative AI analysis unit executes a large-scale language model specifically designed for natural language processing. This large-scale language model employs a Transformer-based decoder architecture with billions of parameters. The generative AI analysis unit uses a tokenizer to segment the input text data into partial strings. The generative AI analysis unit converts each token into a high-dimensional embedding vector. The input data to this model is a three-dimensional tensor composed of the embedding vectors of the tokens. This input data contains positional encodings to preserve contextual information. For example, the first input data contains a vector array transformed from the text "coping slowly". The second input data contains a vector array transformed from the text "explaining carefully". The third input data contains a vector array transformed from the text "feeling uneasy". The output data from this model consists of numerical vectors representing the polarity and intensity of sentiment, and a list of extracted keywords. This output data is structured data that quantifies the intent and sentiment behind the text. For example, the first output data contains a value of 0.9 representing negativity and the keyword "nighttime coping". The second output data contains a value of 0.8 representing positivity and the keyword "eating aid". The third output data contains a value of 0.75 representing the probability of bias. This generative AI analysis unit recalculates the trust score for each manager based on the output sentiment score. If the probability of bias exceeds a predetermined threshold, the generative AI analysis unit instructs subsequent modules to perform corrective processing to ensure the fairness of the evaluation. Through this structure, the system automatically generates objective evaluation indicators from unstructured text. This system eliminates biases in evaluations caused by human subjectivity, significantly improving the transparency and fairness of the evaluation process.This technology can be applied to quality management systems in all service industries that quantify qualitative feedback from customers.

[0084] Based on these analysis results, the evaluation report generation department automatically generates an evaluation report containing specific values ​​and comments such as: "Manager X's turnover rate is 20%, nursing quality score is 4.3, nursing plan achievement rate is 95%, incident rate is 0.2 incidents / person / year, major incident rate is 0.05 incidents / person / year, trust score is 0.75, improvement points: need to strengthen nighttime response, should increase conversations during feeding assistance, thoroughly implement care during elimination assistance, anonymous feedback: 'very detailed explanation,' 'anxious about nighttime response,' 'very detailed feeding assistance,' 'anxious during elimination assistance.'" The report is automatically updated monthly or quarterly and can be viewed by the manager, superiors, and operations headquarters. Furthermore, in the event of a sharp increase in turnover and incident rate, or a sharp decrease in trust score, the alert notification department sends real-time warnings to relevant personnel to prompt early intervention. For example, notifications might include specific situations and recommended actions such as "Employee turnover has exceeded 30% in the past three months. Urgent action is needed," "Numerous anonymous feedback comments expressed 'anxiety about nighttime operations.' A reassessment of the nighttime system is recommended," and "The incident rate has doubled compared to last month. Please re-evaluate employee health and workload." Specifically, this evaluation report generation department uses natural language generation algorithms to dynamically synthesize human-readable text from structured data. This evaluation report generation department executes an in-memory template engine. This evaluation report generation department acquires various scores and extracted keywords from the generative AI analysis department. This evaluation report generation department converts the acquired values ​​into strings according to a prescribed format. This evaluation report generation department collaborates with the alarm notification department to perform event-driven asynchronous processing. The input data for this model is a multi-dimensional key-value data structure containing evaluation indicators for each manager. This input data is an array containing multiple pairs of values ​​and strings. As a specific example, the first input data contains the value 0.2 representing the turnover rate. The second input data contains the value 4.3 representing the care quality score. The third input data contains an array of strings representing improvement points. The output data from this model is text data of a report written in markup language. This output data includes visual embellishments and structured text. For example, the first output data contains a red text block indicating a warning about turnover rates. The second output data contains a dataset depicting the shift in care quality using charts. The third output data contains text blocks with recommended actions written in bullet points. This evaluation report generation unit transmits the output report data to a subsequent distribution server. The alert notification unit immediately generates and sends a data packet for push notifications when specific indicators in the report exceed thresholds. Through this structure, the system instantly transforms massive analytical results into an understandable form, preventing delays in decision-making. This system enables rapid response to abnormal situations within the organization through real-time processing and notification mechanisms.This technology can be applied to business dashboards and crisis management systems that require real-time performance.

[0085] The user interface on the terminal side allows managers, employees, and operations headquarters managers to access the server via a browser or dedicated application to view evaluation reports, publish feedback, conduct evaluation interviews, and confirm improvement proposals. For example, managers can review their own evaluation reports, understand their trust scores and areas for improvement, and thus re-evaluate management actions. Employees can anonymously publish feedback, verify their past feedback history, and track their career growth. Operations headquarters and HR managers can assess the evaluation biases and risks across the entire facility or multiple facilities, and apply this information to system improvements and talent allocation optimization. For example, it provides various interfaces such as dashboards displaying key indicators, detailed displays of individual reports, feedback publication screens, and displays of topics for evaluation interviews. Specifically, this user interface executes a front-end program based on a single-page application architecture. This user interface is built on a virtual document object model running on the client terminal's web browser. This user interface communicates with the server-side application programming interface using an asynchronous communication interface. This user interface stores received report data in an in-memory state management store. This user interface detects state changes and performs high-speed differential rendering of the screen. The input data to this rendering engine is a serialized state object obtained from the server. This input data is a hierarchical data tree. Specifically, the first input data is a node containing the manager's personal information. The second input data is an array of nodes containing a resume with trust scores. The third input data is an array of nodes containing a list of improvement points. The output data from this rendering engine is a collection of pixel information depicted on the screen. This output data represents the configuration information of graphical elements that are visually recognizable to the user. Specifically, the first output data contains a drawing command for a pie chart representing scores. The second output data contains the configuration coordinates of text boxes used for input feedback. The third output data contains a display command for a pop-up window representing an alert. This user interface unit sends the output drawing commands to the graphics processing unit to update the screen. This user interface unit captures input events from the user and triggers subsequent data transmission processing. Through this structure, the system provides smooth operation without perceptible network latency. This system significantly reduces the server's drawing load by effectively utilizing client-side computing resources. This technology can be applied to all network-based business systems involving complex data visualization.

[0086] The feedback input section provides an interface for employees and users to anonymously post feedback to managers and staff. For example, it includes multiple questions and free-response sections such as "Responses at night," "Ease of understanding instructions," "Employee friendliness," "Conversations during eating assistance," "Care during toilet assistance," and "Collaboration among employees," allowing for frank feedback. Posted feedback is immediately sent to the server and analyzed in the generative AI analysis section. For example, it incorporates designs to enhance psychological security, such as a combination of selective questions and free-response input screens, a confirmation screen of past posting history, and automatic response message display after feedback is posted. Specifically, this feedback input section executes an input stream processing module running on the client terminal. It captures character codes entered by the user from the keyboard or touch panel in real time. It sanitizes the input text data to prevent malicious code infiltration. It uses encryption algorithms to convert input data into secure data packets. It sends data to the server via an asynchronous message queue. The input data to this processing section consists of user action events and raw string data. The input data is a one-dimensional array assigned a timestamp and a session identifier. Specifically, the first input data contains an integer value of 3 representing a choice of responses regarding nighttime. The second input data contains a string of free-form data regarding eating assistance. The third input data contains a truth value representing a click event of the send button. The output data from this processing unit is a serialized, encrypted payload that can be sent to a server. This output data is a byte sequence transmitted at the network layer. Specifically, the first output data is a data packet containing header information and encrypted response data. The second output data is a data packet containing a status code indicating input completion. The third output data is a control data packet containing a retransmission request in case of communication errors. This feedback input unit sends the output payload via the network interface. After sending, the feedback input unit clears the input form on the screen and performs subsequent completion message display processing. Through this structure, the system collects data securely and reliably while protecting user privacy. This system completes data transmission in the background without interrupting user input through asynchronous transmission. This technology can be applied to internal notification systems and questionnaire collection systems requiring high anonymity and security.

[0087] The Report Viewing Department provides an interface for managers and operations headquarters heads to view evaluation reports distributed from the server. For example, it displays information such as turnover rate trends over the past year, time-series changes in care quality scores, monthly trends in the number of incidents, heatmaps of bias detection results, trust score trends, and details of improvement point lists in a visually easy-to-understand format. It utilizes various display formats including charts, heatmaps, text summaries, and comparison tables. Specifically, this Report Viewing Department executes a data visualization library that expands on the client terminal's memory. It uses decompression algorithms to restore compressed data received from the server. It performs coordinate transformation processing to map the restored time-series data onto a two-dimensional canvas. It has a mechanism that delays loading only the required data based on the user's scrolling actions. The input data to this visualization engine is a multidimensional array representing the time-series changes of evaluation indicators. This input data contains pairs of timestamps and their corresponding values. As a specific example, the first input data contains an array of floating-point numbers representing the turnover rate for each month. The second input data contains an array of floating-point numbers representing the shift in care quality scores. The third input data contains a two-dimensional matrix representing the bias detection results. The output data from this visualization engine is a set of vector graphics instructions for drawing charts and heatmaps. This output data includes drawing parameters such as color information, line width, and vertex coordinates. As a specific example, the first output data contains instructions to draw the upward trend of the turnover rate with a red line. The second output data contains instructions to draw a heatmap by coloring areas of strong bias with darker colors. The third output data contains instructions to generate tooltips displaying detailed information when the cursor is hovered over a specific data point. The report viewing department passes the output drawing instructions to the browser's rendering pipeline. The report viewing department performs subsequent processing, dynamically recalculating the scale of the charts based on the user's zooming actions. Through this structure, the system transforms complex and multidimensional data into intuitive and understandable visual information. This system reduces the number of communications with the server and improves display speed through dynamic rendering on the client side. This technology can be applied to financial analysis tools and marketing dashboards that process large amounts of statistical data.

[0088] The Assessment Interview Support Department assists managers and subordinates in developing concrete action plans while conducting regular assessment interviews, referencing evaluation reports, feedback records, and AI-generated improvement proposals. For example, it provides AI-generated topics and action lists such as "Analysis of the Causes of Increased Turnover Rate This Period," "Background of Decreased Trust Score," "Specific Countermeasures to Improve Care Quality," "Discussion of Countermeasures to Prevent Recurrence of Incidents," and "Measures to Strengthen Collaboration Among Employees," thereby improving the quality of assessment interviews. It features functions such as automatically listing topics before the interview, displaying past improvement actions and their effects in chronological order, recording notes and action plans in real-time during the interview, and automatically notifying follow-up tasks after the interview. Specifically, this Assessment Interview Support Department uses a real-time status synchronization protocol to share interview status across multiple terminals. It reads the agenda data constructed by generative AI into a state transition machine in memory. It dynamically switches the displayed content based on the progress of the interview. It uses an event sourcing model to record user-inputted notes and action plans as a time-series event log. The input data to this recommendation algorithm consists of feature vectors from past evaluation interview records and current evaluation reports. This input data is a 128-dimensional floating-point vector. Specifically, the first input data contains an embedding vector of text representing the factors contributing to the rising turnover rate. The second input data contains identifiers of past successful improvement measures. The third input data contains numerical values ​​representing the magnitude of the current decrease in care quality scores. The output data from this recommendation algorithm is structured data representing a list of recommended topics and action plans. This output data is an array containing priority scores for each proposal. Specifically, the first output data contains a priority score of 0.9 and the topic "Reassessment of the Nighttime System." The second output data contains a priority score of 0.8 and the topic "Activation of Inter-staff Communication." The third output data contains a priority score of 0.7 and the topic "Reconstruction of the Training Program." The Evaluation Interview Support Department sorts the output list by priority and displays it on the screen. The Evaluation Interview Support Department registers the selected action plans in the subsequent task management database. Through this structure, the system structures the progress of meetings, prevents discussions from deviating from the topic, and enforces the development of effective actions. This system supplements managers' lack of experience by providing data-driven proposals based on past success stories. This technology can be applied to project management tools and online meeting support systems.

[0089] Furthermore, for example, when managers implement improvement actions, the system can report the content from the terminal to the server, and AI can monitor subsequent changes in turnover rate and trust scores to quantitatively evaluate the improvement effect. For instance, the system can automate the verification of the effects of improvement measures such as "as a result of strengthening the nighttime response system, the trust score increased from 0.6 to 0.8," "as a result of increasing conversations during meal assistance, user satisfaction increased from 4.2 to 4.6," and "as a result of implementing measures to strengthen collaboration among employees, the incident rate decreased from 0.2 incidents / person / year to 0.1 incidents / person / year." Specifically, this system executes an effect measurement module that combines a time series forecasting model and a causal inference algorithm. This module retrieves the implementation records of improvement actions input by managers from the database. This module automatically performs statistical hypothesis testing to compare the shifts in evaluation indicators before and after the implementation of the actions. This module generates a control group matched with usage propensity scores to exclude the influence of external factors. The input data to this causal inference model consists of a binary variable representing the presence or absence of the action implementation and a time series array of evaluation indicators during the target period. This input data is represented in the form of a multidimensional tensor. As a specific example, the first input data contains a flag value of 1 indicating the implementation of a strengthened nighttime response system. The second input data contains an array of trust scores for the month prior to implementation. The third input data contains an array of trust scores for the month following implementation. The output data from this causal inference model represents the estimated causal effect of the improvement action on the indicator and its confidence interval. This output data includes a p-value indicating statistical significance. As a specific example, the first output data contains a value of +0.2 indicating the improvement in trust scores. The second output data contains a value of -0.1 indicating the reduction in event occurrence. The third output data contains a p-value of 0.01 indicating the significance of the effect. This module feeds back the estimated causal effect to the evaluation report generation department. This module adds actions that confirm the effect as success cases to the learning data of subsequent recommendation algorithms. Through this structure, the system quantitatively demonstrates the true effect of the policy, not just the correlation. This system uses a feedback loop to allow the AI ​​model to continuously relearn, enabling the accuracy of the proposals to evolve. This technology can be applied to A / B testing of marketing policies and policy evaluation systems.

[0090] As described above, the embodiments of this invention employ a structure where data collection, multi-angle analysis based on generative AI, evaluation report generation, alarm notification, and database management are uniformly performed on the server side, while users on the terminal side can view reports, publish feedback, conduct evaluation interviews, implement improvement actions, and report. This structure enables fair and transparent management evaluation in nursing settings, providing strong support for employee retention, improved nursing quality, increased user satisfaction, and sound organizational operations. Generative AI plays a core role in the system, including comprehensive data analysis, bias detection, feedback analysis based on natural language processing, prediction of the possibility of improved nursing quality, automatic generation of improvement proposals, and verification of the effectiveness of improvement measures. This eliminates the previous subjective evaluations that varied from person to person, enabling a managerial evaluation system that is both objective and transparent. Specifically, this system adopts a microservice architecture, executing each functional module as an independent container on an orchestration platform. The system distributes requests from clients to multiple computing nodes via a load balancer. The system integrates the entire process from data collection to AI analysis and report generation as an automated data pipeline. The system utilizes an in-memory database to maximize the read and write speed of intermediate data. The input data to this integrated control unit consists of status information and log data of processing results asynchronously sent from each microservice. This input data is an array of time-series metrics representing the overall system's operational status. Specifically, the first input data includes event notifications indicating the completion of processing by the data collection unit. The second input data includes values ​​representing the inference latency of the AI ​​analysis unit. The third input data includes values ​​representing the memory usage of the evaluation report generation unit. The output data from this integrated control unit consists of control commands instructing the dynamic allocation of system resources. This output data is a set of instructions for performing horizontal scaling of containers. Specifically, the first output data includes commands to add and start computational instances for AI analysis. The second output data includes commands to release unnecessary cache memory. The third output data includes commands instructing traffic to be switched from failed nodes. This integrated control unit sends the output commands to the infrastructure management interface to dynamically change the system architecture. This integrated control unit performs follow-up processing to autonomously eliminate processing bottlenecks. Through this architecture, the system maintains high throughput and fault tolerance even under bursty access concentrations and large-scale data input. This system maximizes the utilization efficiency of computer resources and reduces operating costs while providing a stable, objective, and transparent evaluation basis. This technology can be applied to overall enterprise cloud services for mission-critical applications.

[0091] (System Composition) The system described in this embodiment includes a data collection department, a database management department, a generative AI analysis department, an evaluation report generation department, an alarm notification department, a user interface department, a feedback input department, a report viewing department, and an evaluation interview support department. The data collection department automatically and regularly collects data about managers and subordinate employees from various information sources both inside and outside the nursing facility. For example, it obtains data from the nursing staff's personnel management system, such as hiring date, termination date, transfer intentions, job transfer history, training history, qualification status, years of continuous service, past evaluation scores, promotion history, and reward / punishment history. It also obtains data from the nursing record system, such as nursing plan achievement rate, daily nursing records, nursing content for each user, nursing implementation frequency, details of nursing content (eating assistance, bathing assistance, excretion assistance, rehabilitation support, recreation implementation status, etc.), user satisfaction questionnaire results, each user's nursing satisfaction score, and records of direct thanks and complaints from users. Finally, it obtains data from the incident / near-incident reporting system, such as the number of incidents, date and time of occurrence, location of occurrence, cause of occurrence, response measures, and prevention of recurrence. The data collection department collects information such as countermeasures, participating employees, severity classification of incidents, and follow-up status after incidents. It also gathers information on employee communication records, including business contact systems and chat logs, meeting attendance records, business handover records, employee consultation / suggestion records, meeting minutes of regular meetings, and exchanges of thank-you / praise messages. Furthermore, it collects information on stress levels, health risk scores, doctor's guidance, implementation status of health maintenance activities, and mental health counseling records from regular stress checks and health diagnoses. Finally, it collects anonymous feedback and 360-degree feedback from users and their families, including opinions based on free writing, scores based on 5-stage and 10-stage evaluations, individual comments on specific employees, letters and emails from users' families, thank-you notes and complaints, feedback submission frequency, and time-series changes in feedback. Specifically, this data collection department generates and executes parallel processing threads on a computing server with multi-core processors. This department constructs a data pipeline responsible for extraction, transformation, and storage. This department normalizes heterogeneous and mixed data from diverse external information sources into a unified internal format. This data collection department applies data cleaning algorithms to impute missing values ​​and remove outliers. The input data to this preprocessing module consists of multiple raw data records with different patterns. This input data is a one-dimensional array containing various types of data, including strings, integers, floating-point numbers, and dates. For example, the first input data contains a date string representing the hiring date obtained from the personnel management system. The second input data contains freely written text data obtained from the nursing record system. The third input data contains numerical data representing health diagnosis results. The output data from this preprocessing module is a multidimensional numerical tensor converted into a form that can be processed by a machine learning model. This output data includes data where categorical variables have been converted into binary arrays.As a specific example, the first output data contains a floating-point number of 0.5 normalized to the number of consecutive working years. The second output data contains a 512-dimensional array vectorized from the text. The third output data contains integer values ​​representing health risks in three stages. This data collection unit transmits the output numerical tensors to the subsequent database management unit via a high-speed message queue. This data collection unit asynchronously writes the processing completion status to a log file. Through this structure, the system integrates massive amounts of data in different formats in real time, ensuring the accuracy of subsequent AI analysis. This system significantly reduces the computational cost and time required for data preparation. This technology can be applied to the foundation of digital transformation by integrating multiple legacy systems.

[0092] The Database Management Department organizes and manages the diverse data acquired by the Data Collection Department according to time series, enabling retrieval, extraction, and summarization from multiple perspectives, including individual managers, employees, users, departments, facilities, and events. For example, it builds a data foundation for detailed analysis, such as the turnover rate of a specific manager's subordinates over the past three years, annual comparisons of transfer intention rates for a particular facility, time-series changes in the nursing quality score of a specific employee, monthly trends in the number of incidents, time-series changes in anonymous feedback, correlation analysis between training experience and nursing quality score, the relationship between employee communication frequency and incident rate, the correlation between user satisfaction and nursing plan achievement rate, the relationship between employee health risk score and turnover rate, and the relationship between the frequency of feedback from family members and the number of complaints. For instance, it can also extract periods of high transfer intention rates for each facility and compare them with organizational restructuring timing, extract deviations and areas of expertise in each employee's nursing content, and utilize this information for appropriate talent allocation and training plan development. Specifically, this Database Management Department employs a hybrid structure combining relational and non-relational databases. This database management department dynamically allocates routing engines to store destinations based on the characteristics of the data. To achieve high-speed read and write of large-scale data, this department applies sharding technology, distributing data across multiple physical servers. This department uses a query optimizer to optimize the execution plan for retrieval requests. The input data to this search engine is a multi-dimensional retrieval query issued by the user or within the system. This input data is a structured set of parameters specifying search criteria, sorting order, number of retrieved items, etc. For example, the first input data contains parameters specifying the past three years for a particular facility. The second input data contains a threshold condition such as a turnover rate exceeding a certain value. The third input data contains a sorting condition of descending order of care quality scores. The output data from this search engine is a set of records that meet the criteria, along with statistical values ​​calculated by a summary function. This output data is an array of pointers to the result set expanded in memory. For example, the first output data contains a list of identifiers for employees who meet the criteria. The second output data contains a numerical value representing the average number of events occurring per month. The third output data contains metadata representing the execution time required for retrieval processing. This database management department returns the output result set to the module that requested the request. This database management department performs subsequent processing by storing the results of frequently executed queries in a memory cache. Through this architecture, the system guarantees millisecond-level response times even for petabyte-level data. By completing complex aggregation processing on the database side, the system minimizes network bandwidth consumption. This technology can be applied to large-scale log analysis systems and real-time monitoring systems.

[0093] The Generative AI Analytics Department is responsible for performing diverse AI analyses using data stored in the Database Management Department, combining large-scale language models, gradient boosting models, random forests, neural networks, natural language processing models, clustering models, anomaly detection models, and more. The generative AI analytics department calculates the turnover rate and transfer intention rate of each manager's subordinates. For example, it summarizes the number of employees who left and the number of employees who expressed transfer intentions in the past year, divides them by the total number of subordinates, and uses this as a quantitative indicator. Regarding nursing quality, it comprehensively evaluates nursing plan achievement rate, user satisfaction questionnaires, training experience, qualification status, diversity of nursing content, frequency of nursing implementation, nursing satisfaction score for each user, and records of user thanks / complaints, etc. For example, it outputs multiple indicators such as "the average nursing quality score of subordinates is 4.3 points", "the nursing plan achievement rate is 95%", "the average user satisfaction score is 4.5 points", and "the diversity score of nursing content is 0.8". Regarding accidents / near-accidents, it calculates the number of reports per employee and obtains detailed indicators such as "the incident occurrence rate is 0.2 incidents / person / year", "the major incident occurrence rate is 0.05 incidents / person / year", and "the recurrence rate of incidents is 10%". Furthermore, regarding anonymous feedback from users and their families, generative AI uses natural language processing to analyze the text, extracting keywords and emotions such as "trustworthy," "friendly," "detailed explanations," "slow response," "feeling uneasy," "dissatisfaction with nighttime responses," "employee smiles provide reassurance," "detailed mealtime assistance," and "insufficient care during toileting assistance." It then creates a trust score and a list of improvement points for each manager. In cases with many negative feedback, it automatically generates specific improvement proposals such as "insufficient explanations from employees," "anxiety about nighttime responses," "few conversations during mealtime assistance," and "insufficient care during toileting assistance." It analyzes the distribution of employee evaluation data and the tendencies of each attribute, detecting bias risks such as "consistently giving low ratings to female employees," "giving high ratings to specific qualification holders," "strict ratings to newly hired employees," and "lenient ratings to long-term continuous workers," calculating bias scores. It also analyzes the relationship between employee health risk scores and stress test results and turnover and incident rates, extracting causal relationships such as "employees with high stress scores have a 1.5 times higher turnover rate than average" and "employees with high health risks have a 2 times higher incident rate."Specific examples of prompts that the generative AI can read include: "Calculate the turnover rate, transfer intention rate, nursing quality score, and incident rate of manager X's subordinates over the past year; conduct sentiment analysis of anonymous feedback to extract trust scores and improvement points"; "Predict the likelihood of improved nursing quality in the next six months based on employee Y's nursing records, training history, and satisfaction questionnaire results"; "Analyze manager Z's employee evaluation data by each attribute, detect any evaluation bias, and calculate bias scores"; and "Analyze anonymous feedback text data through natural language processing, summarize negative opinions, and generate improvement proposals." Specifically, this generative AI analysis department uses prompting engineering techniques to control the inference process of a large-scale language model. This generative AI analysis department loads a specialized model, fine-tuned for specific tasks, into memory. This generative AI analysis department has a learning phase that uses a cross-entropy error function to calculate the difference between the model's output and the correct answer data, and updates the weight parameters using gradient descent. During the inference phase, this generative AI analysis department rapidly calculates the probability distribution for the input data. The input data to this model is an integer array of tokenized prompt strings automatically generated by the system. The input data includes contextual information and specific instructions. For example, the first input data includes a labeled column indicating the calculation of manager X's turnover rate. The second input data includes a labeled column indicating the prediction of the likelihood of improved care quality. The third input data includes a labeled column indicating the detection of evaluation bias. The output data from this model consists of a labeled column of generated text and an array of values ​​representing the generation probability of each labeled column. The output data includes a natural language-based response and a score representing the confidence level of that response. For example, the first output data includes a labeled column of text indicating a turnover rate of 20%. The second output data includes a labeled column of text indicating a bias score of 0.8. The third output data includes a value representing the generation probability of 0.95. This generative AI analysis unit uses a decoder to inversely convert the output labeled columns into human-readable strings. If the confidence score falls below a predetermined threshold, the generative AI analysis unit performs subsequent processing to discard the inference result and request recalculation. Through this structure, this system flexibly performs complex analytical tasks using only natural language instructions. This system can understand the context and derive appropriate results even for fuzzy inputs that rule-based programs cannot handle. This technology can be applied to automated response systems and decision support systems that require a high level of reasoning ability.

[0094] The Evaluation Report Generation Department automatically generates evaluation reports for each manager, including data such as turnover rate, transfer intention rate, nursing quality score, nursing plan achievement rate, incident rate, major incident rate, trust score, list of improvement points, and bias detection results obtained from the Generative AI Analysis Department. The reports are automatically updated monthly or quarterly and can be viewed by the manager, their superiors, and the operations headquarters. For example, the system automatically generates evaluation reports containing specific figures and comments such as "Manager X's turnover rate is 20%, care quality score is 4.3, care plan achievement rate is 95%, incident rate is 0.2 incidents / person / year, major incident rate is 0.05 incidents / person / year, trust score is 0.75, improvement points: need to strengthen nighttime response, increase conversations during meal assistance, thoroughly implement care during toilet assistance, anonymous feedback: 'detailed explanation,' 'anxiety about nighttime response,' 'detailed meal assistance,' 'anxiety about toilet assistance,'" etc. Furthermore, in cases of a sharp increase in turnover and incident rate, or a sharp decrease in trust score, the alert notification department sends real-time warnings to relevant personnel to prompt early intervention. For example, notifications may include: "Subordinate employee turnover has exceeded 30% in the past 3 months. Urgent response is needed," "Most anonymous feedback received comments of 'anxiety about nighttime response.' A reassessment of the nighttime system is recommended," and "Incident rate has doubled compared to last month. Please reassess employee health and workload," etc. Specifically, this evaluation report generation unit is executed as a background process periodically started by a batch processing scheduler. This unit expands a markup language and document format generation engine in memory. It maps structured data received from the generative AI analysis unit into predefined layout templates. It constructs a tree structure of a document object model in memory. The input data to this generation engine is multidimensional, typical data containing evaluation metrics for each manager, a list of improvement points, and bias detection results. This input data is a collection of key-value pairs. For example, the first input data contains a pair of the key "turnover rate" and the value "20 percent". The second input data contains a pair of the key "trust score" and the value "0.75". The third input data contains an array of key "improvement points" and their values. The output data from this generation engine is serialized into a renderable binary or text file. This output data includes font information, layout coordinates, and embedded image data. For example, the first output data contains a byte column of a document file embedding chart images. The second output data contains a text string that can be displayed in a web browser. The third output data contains metadata indicating the generation date and time and file size. The evaluation report generation department saves the output file in a secure storage area. This department then performs subsequent processing, including recording the file's save path in the database and sending a completion signal to the alarm notification department.This architecture enables the system to generate personalized reports for thousands of managers in a short period of time. The system completely eliminates the time spent on manual report creation, maintaining the freshness of information. This technology can be applied to monthly reporting systems in financial institutions and other organizations that require a large number of form outputs.

[0095] The User Interface (UI) department is used by managers, employees, and operations headquarters managers to access the server via browsers and dedicated applications to view evaluation reports, publish feedback, conduct evaluation interviews, and confirm improvement proposals. For example, managers can review their own evaluation reports, understand their trust scores and areas for improvement, and reassess management actions. Employees can anonymously publish feedback, review past feedback records, and track their career growth. Operations headquarters and HR managers can assess evaluation biases and risks across the entire facility and multiple facilities, applying this information to system improvements and talent optimization. It provides various interfaces, including dashboards displaying key metrics, detailed individual reports, feedback submission screens, and evaluation interview topic displays. Specifically, this UI department manages the virtual document object model of the front-end framework running on client terminals. It asynchronously captures user action events and updates the data stored in the status management system. It communicates with the back-end server via a communication interface gateway. This UI department implements a role-based access control mechanism, dynamically generating screen displays based on user permissions. The input data to this access control module consists of the user's authentication token and the identifier of the requested resource. This input data includes encrypted strings and path information. Specifically, the first input data includes an authentication token representing administrator privileges. The second input data includes a path string requesting access to a report from a specific facility. The third input data includes the user's session identifier. The output data from this access control module consists of a true / false value for access permission and filtering rules for the data allowed to be displayed to the user. This output data is a set of parameters used to control subsequent data retrieval processing. Specifically, the first output data includes a true / false value representing access permission. The second output data includes a conditional expression indicating that only data from this department will be retrieved. The third output data includes a flag indicating that the edit button should be hidden. This user interface unit assigns the output filtering rules and sends a data retrieval request to the server. This user interface unit performs subsequent processing, binding the received data to components on the screen and redrawing it. Through this structure, the system securely and appropriately provides information to diverse user layers through a single application. This system robustly prevents unauthorized access to data beyond authorized limits on both the client and server sides. This technology can be applied to internal portal systems in companies with complex organizational hierarchies.

[0096] The Feedback Input Section provides an interface for employees and users to anonymously post feedback to managers and staff. It includes multiple questions and open-ended fields such as "Responses to Nighttime Events," "Ease of Understanding of Explanations," "Employee Friendliness," "Conversations During Eating Assistance," "Caring During Toilet Assistance," and "Collaboration Among Employees," allowing for frank feedback. Posted feedback is immediately sent to the server and analyzed by the Generative AI Analysis Section. Measures to enhance psychological security are implemented, including a combination of selective questions and open-ended input screens, a confirmation screen for past postings, and automatic reply messages after feedback is posted. Specifically, this Feedback Input Section keeps an event listener monitoring the input form's status resident in the client terminal's memory. It performs real-time validation of input values ​​using regular expressions each time the user enters text. It employs escaping techniques to prevent malicious script attacks. It temporarily stores data during input in the browser's local storage in case of unexpected communication interruptions. The input data to this validation module consists of the input field identifier and the user's raw string data. This input data is an array of character codes extracted from keyboard events. Specifically, the first input data contains identifiers for questions about nighttime responses. The second input data contains a text string entered by the user. The third input data contains an integer value indicating the selection state of a radio button. The output data from this verification module consists of a true / false value indicating the appropriateness of the input value and a warning message string indicating an error. This output data is a control signal used to update the display state of the screen. Specifically, the first output data contains a true / false value indicating appropriateness. The second output data contains an error message indicating that the character limit has been exceeded. The third output data contains a warning flag indicating that a required field has not been entered. This feedback input unit only enables the send button if the output true / false value is true. When the send button is pressed, this feedback input unit performs subsequent processing to serialize the input data into a structured format and send it asynchronously to the server. Through this structure, the system proactively prevents incomplete or improper data from reaching the server. This system reduces input errors and improves the quality of data collection by providing immediate feedback to the user. This technology can be applied to online application systems that require high data consistency.

[0097] The Report Viewing Department provides managers and operations headquarters executives with an interface to view evaluation reports distributed from the server. It displays information such as turnover rate trends over the past year, time-series changes in care quality scores, monthly trends in the number of incidents, heatmaps of bias detection results, trust score trends, and details of improvement point lists in a visually easy-to-understand format. It utilizes various display formats including charts, heatmaps, text summaries, and comparison tables. Specifically, this Report Viewing Department uses a web graphics library to optimize the hardware acceleration of the client-side graphics processing unit. This Report Viewing Department executes shader programs for high-speed drawing of charts with large amounts of data points. This Report Viewing Department implements a culling process that skips the drawing of elements located off-screen. This Report Viewing Department deploys a cross-filtering engine in memory that dynamically filters data based on user interaction. The input data to this filtering engine consists of the filter conditions selected by the user on the screen and the entire dataset cached in memory. This input data is a combination of multidimensional arrays and conditional expressions. As a specific example, the first input data contains period filter conditions specifying a particular month. The second input data contains category filtering conditions specifying particular event types. The third input data contains a multidimensional array of evaluation scores for all employees. The output data from this filtering engine is an array of subsets containing only the data points that meet the conditions. This output data uses a lightweight data structure optimized for re-rendering. For example, the first output data contains an array containing only data for a specified month. The second output data contains a set of vectors representing the coordinates of the occurrence of a specific event. The third output data contains a recalculated value representing the average score after filtering. The report viewing department passes the output subset array to the rendering pipeline, synchronously updating all charts on the screen. The report viewing department saves the filtering status as a query parameter in the URL and performs subsequent processing that enables status sharing. Through this structure, the system enables complex dashboards containing tens of thousands of data points to operate without delay. This system supports users in exploring data from multiple perspectives, promoting deeper insights. This technology can be applied to business intelligence tools that process big data.

[0098] The Performance Appraisal and Interview Support Department assists managers and subordinates in developing concrete action plans during regular performance appraisal and interview sessions. This is achieved by referring to evaluation reports, feedback documents, and AI-based improvement proposals. The department provides AI-generated lists of topics and actions, such as "Analysis of the Causes of Increased Turnover Rate This Period," "Background of Decreased Trust Score," "Specific Countermeasures to Improve Care Quality," "Discussion of Countermeasures to Prevent Recurrence of Incidents," and "Measures to Strengthen Collaboration Among Employees," thereby improving the quality of performance appraisal and interviews. Features include automatically listing topics before the interview, displaying past improvement actions and their effects in chronological order, real-time note-taking and action planning during the interview, and automatic follow-up notifications after the interview. Specifically, the department manages the state machine of the interview process on a server. It uses a two-way communication protocol to synchronize screen status and input content between participants' terminals in milliseconds. The department uses a natural language generation model to automatically construct the optimal agenda from past data and current issues. After the interview, the department performs text analysis to extract key decisions. The input data to this text analysis module consists of text data from meeting minutes recorded during interviews and input logs of action plans. This input data is an array of timestamped strings. Specifically, the first input data contains metadata indicating who spoke and when. The second input data contains text of the specific countermeasures decided. The third input data contains strings indicating the person in charge and the deadline. The output data from this text analysis module is the extracted structured data of the tasks. This output data is an object in a format that can be registered in a task management system. Specifically, the first output data contains a reassessment of the task name and the nighttime system. The second output data contains a pair of person in charge identifiers and deadline dates. The third output data contains an integer value indicating the priority of the task. The interview support department inserts the output task data into the database. The interview support department performs follow-up processing by sending push messages to the person in charge's terminal to notify them of the task assignment. Through this structure, the system fully automates the conversion from meeting minutes to actionable tasks. This system prevents omissions of decisions and dramatically improves organizational execution. This technology can be applied to retrospective support tools in agile development, etc.

[0099] In this embodiment, the system achieves data collection, data management, AI analysis, report generation, alarm notification, user interface, feedback input, report viewing, and evaluation interview support through the coordinated operation of various components. It uses generative AI as its core to improve the fairness, transparency, and objectivity of management evaluations in nursing settings, thereby achieving employee stability, improved nursing quality, increased user satisfaction, and sound organizational operation. Specifically, this system adopts an event-driven architecture where the aforementioned components are loosely coupled via asynchronous message queues. The system positions generative AI as the core inference engine, constructing a pipeline through which all data flows via the AI ​​model. The system models a closed-loop control system from data input to analysis, report generation, and feedback collection. Through a pipeline of continuous integration and continuous deployment, the system periodically updates the weight parameters of the AI ​​model. The input data to this inference engine consists of multi-dimensional usage metrics and user action logs collected from the entire system. This input data is a large-scale time-series tensor with millions of rows. For example, the first input data contains a numerical array representing the number of calls to each functional module. The second input data contains a numerical array representing the user's screen dwell time. The third input data contains a numerical array representing the shift in the error rate of the prediction model. The output data from this inference engine consists of adjusted parameters for system-wide optimization and future performance predictions. This output data is a control vector passed to the self-tuning algorithm. Specifically, the first output data contains adjustments to extend the cache's expiration time. The second output data contains flags indicating relearning of specific AI models. The third output data contains a numerical value representing the system load prediction for next week. The system uses the output control vectors to autonomously optimize its system architecture. The system performs subsequent processing by pre-configuring server resources based on the predicted load. Through this architecture, the system achieves a self-evolving platform where evaluation accuracy and system processing efficiency improve with continued use. The system guarantees a high degree of objectivity and transparency while minimizing human intervention. This technology can be applied to next-generation social infrastructure systems that dynamically model complex human societies and propose optimal interventions.

[0100] (Implementation steps) Step 1: Comprehensive collection of nursing field data In this step, the data collection department automatically and regularly collects data on managers and their subordinates from a variety of information sources both inside and outside the nursing facility. For example, it obtains data from the nursing staff's personnel management system, including hiring date, termination date, transfer intentions, job transfer history, training history, qualification status, years of continuous service, past performance evaluations, promotion history, and reward / punishment history. It also obtains data from the nursing record system, including nursing plan achievement rate, daily nursing records, nursing content for each user, frequency of nursing implementation, details of nursing content (eating assistance, bathing assistance, excretion assistance, rehabilitation support, recreational implementation status, etc.), user satisfaction questionnaire results, each user's nursing satisfaction score, and records of direct thanks and complaints from users. The data collection department obtains information from the accident / near-miss incident reporting system, including the number of incidents, date and time of occurrence, location, cause, response measures, recurrence prevention strategies, participating employees, severity classification, and follow-up status. This information is used as employee communication history, including business contact system and chat logs, meeting attendance records, business handover records, employee discussion / suggestion records, meeting minutes of regular meetings, and employee thank-you / praise messages. As a result of regular stress checks and health diagnoses, the department also obtains stress levels, health risk scores, guidance from doctors, implementation status of health maintenance activities, and mental health consultation records. Furthermore, it collects anonymous feedback and 360-degree feedback from users and their families, including freely filled opinions, scores based on 5-stage or 10-stage evaluations, individual comments on specific employees, letters and emails from users' families, thank-you notes and complaints, feedback submission frequency, and time-series changes in feedback, all from questionnaires and dedicated feedback systems. Specifically, this data collection department operates in a distributed computing environment. It uses asynchronous communication protocols to establish connections with multiple external systems. This data collection unit acquires data via application programming interfaces (APIs) provided by various systems. The acquired data is buffered in a temporary memory area. Data cleaning is performed on the buffered data. Statistical methods are used to supplement or remove missing and outlier values. The cleaned data is converted into a unified data model. The converted data is then transmitted to a time-series database. Data compression algorithms are applied during data transmission. Network bandwidth consumption is minimized. Regular batch processing and real-time stream processing are performed in parallel. Data freshness is maintained at all times. Encrypted communication protocols are used to protect data confidentiality. Access control lists are used to verify data access permissions. Unauthorized access is blocked. The overall system processing load is monitored. The data acquisition frequency is dynamically adjusted based on load conditions.This data collection department automates data collection through these processes. It completely eliminates manual data entry based on human intervention. It prevents human error during data entry. It significantly reduces the time required for data collection. It improves the processing efficiency of the entire computer system. It enables real-time monitoring in large-scale nursing facility networks. It provides a unified foundation for managing information from diverse data sources.

[0101] Step 2: Time-series storage and multi-axis management in the database In this step, the Database Management Department organizes and manages the diverse data collected in the previous step according to time series, enabling retrieval, extraction, and summarization from multiple perspectives, such as individual managers, employees, users, departments, facilities, and events. For example, it builds a data foundation for detailed analysis, including trends in employee turnover rates under specific managers over the past three years, annual comparisons of transfer intention rates for a particular facility, time-series changes in nursing quality scores for specific employees, monthly trends in the number of incidents, time-series changes in anonymous feedback, correlation analysis between training experience and nursing quality scores, the relationship between employee communication frequency and incident incidence, the correlation between user satisfaction and nursing plan achievement rates, the relationship between employee health risk scores and turnover rates, and the relationship between the frequency of feedback from family members and the number of complaints. Specifically, the Database Management Department implements highly available data storage built on a distributed file system. It uses an index structure specifically designed for time-series data to store the collected data. It builds multidimensional indexes for high-speed processing of multidimensional retrieval queries. The Database Management Department represents the data as nodes and edges in a graph database. This database management department defines managers, employees, users, and events as nodes. It defines the relationships and interactions between them as edges. This department executes graph exploration algorithms. It extracts complex relationships at high speed in high-dimensional space. It utilizes in-memory databases. It caches frequently accessed data in memory. It reduces disk I / O operations. It minimizes data retrieval latency. It performs data redundancy processing. It prevents data loss in the event of hardware failure. It performs data partitioning. It distributes large-scale data across multiple storage nodes. It enables parallel processing-based aggregation operations. These processes dramatically improve retrieval speed. It performs unconventional steps fundamentally different from traditional human-based spreadsheet software aggregation. It discovers hidden data relationships through operations in high-dimensional space. It enables comprehensive analysis of personnel and care quality data across multiple facilities. This database management department provides the data foundation for immediate response to ad-hoc data analytics requirements from management.

[0102] Step 3: Multi-faceted evaluation / analysis based on generative AI In this step, the Generative AI Analytics Department uses data accumulated in the Database Management Department to perform various AI analyses. It calculates the turnover rate and transfer intention rate of each manager's subordinates, for example, summarizing the number of employees who left and the number who expressed transfer intentions in the past year, dividing by the total number of subordinates to obtain quantitative indicators. Regarding nursing quality, it comprehensively evaluates nursing plan achievement rate, user satisfaction questionnaires, training attendance records, qualification status, diversity of nursing content, frequency of nursing implementation, nursing satisfaction scores for each user, and records of user thanks / complaints, etc. Regarding incidents / near misses, it calculates the number of reports per employee, the incidence rate of major incidents, and the recurrence rate, etc. Furthermore, regarding anonymous feedback from users and their families, generative AI uses natural language processing to analyze the text, extracting keywords and sentiments such as "trustworthy," "friendly," "careful explanation," "slow response," "feeling uneasy," "dissatisfaction with nighttime responses," "employee smiles provide reassurance," "careful feeding assistance," and "insufficient care during toileting assistance." It then creates trust scores and improvement point lists for each manager, analyzes the distribution of employee evaluation data and the tendency of each attribute, and also analyzes the relationship between bias risk, health risk scores, turnover rate, and incident incidence rate. Specific examples of prompts for the generative AI to read include: "Calculate the turnover rate, transfer intention rate, nursing quality score, and incident incidence rate of manager X's subordinates over the past year; conduct sentiment analysis on anonymous feedback to extract trust scores and improvement points," "Based on employee Y's nursing records, training history, and satisfaction questionnaire results, predict the likelihood of improved nursing quality in the next six months," "Analyze manager Z's employee evaluation data by each attribute, detect any evaluation bias, and calculate bias scores," and "Analyze anonymous feedback text data through natural language processing, summarize negative opinions, and generate improvement proposals," etc. Specifically, this generative AI analysis unit executes multiple neural networks including large-scale language models and natural language processing models. This generative AI analysis unit employs a deep learning model based on the Transformer architecture. This generative AI analysis unit receives word embedding vectors from text data as input data. For example, this generative AI analysis unit processes 768-dimensional floating-point vectors as input. As a specific example of input data, this generative AI analysis unit receives tensors vectorized from the text "slow response" and tensors vectorized from the text "employees' smiles provide reassurance." This generative AI analysis unit receives one-dimensional tensors normalized to turnover rate and incident occurrence rate as numerical data as input. As a specific example of input data, this generative AI analysis unit receives normalized values ​​representing a turnover rate of 0.2 and normalized values ​​representing an incident occurrence rate of 0.05. This generative AI analysis unit maps these input data into a high-dimensional feature space. This generative AI analysis unit uses a self-attention mechanism to compute contextual dependencies between data.This generative AI analysis unit generates trust scores and sentiment classification labels as output data. The output data format includes a score represented by floating-point numbers from 0.0 to 1.0, and a probability distribution vector representing positive, negative, and neutral values. As a specific example of the output data, this generative AI analysis unit outputs a trust score of 0.85 and a negative probability of 0.92. The output results are then passed to subsequent processing. If the score falls below a pre-set threshold, a warning flag is generated. This flag is sent to the evaluation report generation unit and the warning notification unit. This generative AI analysis unit performs self-supervised learning and fine-tuning as the model's learning methods. The generative AI analysis unit uses a cross-entropy error function to minimize prediction errors. The generative AI analysis unit uses backpropagation to update the network's weight parameters. These processes eliminate subjective human evaluations. This generative AI analysis unit achieves high-precision sentiment extraction and objective calculation of evaluation metrics. This generative AI analytics department leverages the computing power of computers to instantly extract useful information from massive amounts of unstructured data. This generative AI analytics department brings about a significant improvement in the accuracy and processing speed of personnel evaluations.

[0103] Step 4: Automatic generation and distribution of evaluation reports In this step, the evaluation report generation department automatically generates an evaluation report for each manager, including data obtained from the generative AI analysis department, such as turnover rate, transfer intention rate, nursing quality score, nursing plan achievement rate, incident rate, major incident rate, trust score, list of improvement points, and bias detection results. The reports are automatically updated monthly or quarterly and can be viewed by the manager, their supervisor, and operations headquarters. For example, the automatically generated report might include specific values ​​and comments such as: "Manager X's turnover rate is 20%, nursing quality score is 4.3, nursing plan achievement rate is 95%, incident rate is 0.2 incidents / person / year, major incident rate is 0.05 incidents / person / year, trust score is 0.75, improvement points: need to strengthen nighttime response, increase conversation during feeding assistance, thoroughly implement care during elimination assistance, anonymous feedback: 'careful explanation,' 'anxiety about nighttime response,' 'careful feeding assistance,' 'anxiety during elimination assistance.'" Specifically, this evaluation report generation department executes a natural language generation algorithm. This evaluation report generation department receives structured data output from the generative AI analysis department. It converts the structured data into unstructured text. The department maintains multiple predefined report templates in memory. It dynamically selects the optimal template based on the attributes of the input data. It inserts numerical data and extracted text into placeholders in the selected template. It executes a dynamic hypertext markup language generation engine. It constructs report data in a format displayable by a web browser. It utilizes a data caching mechanism during report generation. It eliminates repetitive generation processes for the same data. It significantly reduces computational resource consumption. It distributes the generated report data via a secure communication channel. It implements encryption, allowing only authorized users to decrypt the report. These processes completely eliminate the human cost required for report creation. It eliminates information delays based on real-time distribution. This makes it possible to consistently display the latest evaluation data on automated dashboards during monthly management meetings and daily business management. This evaluation report generation department, through computer-based automation, not only improves the efficiency of business processes but also enhances the accuracy and timeliness of information delivery.

[0104] Step 5: Early detection and response to risks based on alert notifications In this step, the Alert Notification Department sends real-time alerts to managers, operations headquarters, and management when it detects significant risks such as a sharp increase in turnover or incident rate, or a sharp decrease in trust score, prompting early response. For example, alerts may include specific situations and recommended actions such as "Employee turnover has exceeded 30% in the past three months. Urgent action is needed," "Numerous anonymous feedback comments have expressed 'anxiety about nighttime operations.' A reassessment of the nighttime system is recommended," and "The incident rate has doubled compared to last month. Please re-evaluate employee health and workload." Specifically, this Alert Notification Department executes anomaly detection algorithms. It uses machine learning models such as Isolation Forest and Single-Class Support Vector Machine. The department receives a time-series array of scores and a vector representing the number of incidents as input data. As a specific example of input data, the department receives a 30-dimensional vector representing the number of incidents in the past 30 days and a time-series tensor representing the daily shift in trust score. The department then compares the input data with the distribution of normal data. This alert notification unit determines whether the input data deviates statistically significantly from the normal distribution. It generates an anomaly score and an alert flag as output data. The output data format includes an anomaly score ranging from 0.0 to 1.0 and an integer value representing the importance level. A specific example of the output data is an anomaly score of 0.95 and an importance level of 3. The unit initiates a push notification protocol when the anomaly score exceeds a pre-set threshold. It uses WebSocket communication to send notification messages to client terminals in real time. It collaborates with an email server to distribute emergency emails. These processes enable early detection of significant risks. The unit plays a crucial role in preventing incidents. It reduces system response time, providing an environment where managers can take immediate action. It functions as a precursor detection system for major incidents. Through computer technology, it achieves continuous 24 / 7 / 365 monitoring, exceeding the limits of human surveillance capabilities.

[0105] Step 6: Visualize and apply information through the user interface. In this step, the user interface provides an interface for managers, employees, operations headquarters managers, and other users to access the server via a browser or dedicated application to view evaluation reports, publish feedback, conduct evaluation interviews, and confirm improvement proposals. Managers can review their own evaluation reports, understand trust scores and improvement points to re-evaluate management practices. Employees can anonymously publish feedback, review past feedback history and their career development. Operations headquarters and HR managers can understand the evaluation biases and risks across the entire facility or multiple facilities and apply them to system improvements and talent allocation optimization. Specifically, this user interface implements the front-end rendering engine. This user interface uses asynchronous data acquisition technology to retrieve the latest data from the server. This user interface dynamically updates portions of the screen without reloading the entire page. This user interface reduces the rendering load on client terminals. This user interface minimizes network traffic. This user interface implements a graphics rendering library. This user interface converts time-series and comparative data into visually easy-to-understand charts and graphs. This user interface implements a role-based access control mechanism. This user interface dynamically restricts the information displayed and available functions based on the logged-in user's permission level. This user interface employs a secure token authentication method. This user interface prevents session hijacking and unauthorized access. This user interface utilizes responsive design. This user interface provides optimal layouts for devices with various screen sizes, including PCs, tablets, and smartphones. Through these processes, this user interface significantly improves the visual readability and operability of information. This user interface provides users with an environment that allows them to intuitively analyze data and make decisions. This user interface addresses various use cases, such as real-time feedback viewing in smartphone applications and comprehensive dashboard operations for management. This user interface brings technological improvements to optimize the interaction between computers and humans.

[0106] Step 7: Feedback Release and Evaluation Interview Implementation. In this step, the Feedback Input Department accepts anonymous feedback from employees or users. The Evaluation Interview Support Department supports managers in developing specific action plans while conducting regular evaluation interviews with their subordinates, referencing evaluation reports, feedback history, and AI-based improvement proposals. Features include automatically listing topics before the interview, displaying past improvement actions and their effects in chronological order, recording notes and action plans in real-time during the interview, and automatically notifying follow-up tasks after the interview. Specifically, the Feedback Input Department performs real-time cleansing of the input text data. It escapes special characters to prevent cross-site scripting and SQL injection attacks. It uses a natural language processing model to automatically add tags to the input text. It extracts keywords from the text and automatically assigns relevant category labels. The Evaluation Interview Support Department executes a group state machine to control the interview process. It asynchronously retrieves past evaluation reports and feedback history from the database. The Evaluation Interview Support Department displays the retrieved data in chronological order on the screen. This assessment and interview support department displays improvement proposals generated by the generative AI analysis department as prompts on the screen. This assessment and interview support department synchronizes interview notes and action plans to the server in real time. This assessment and interview support department uses optimistic concurrency control to resolve conflicts caused by multiple users editing simultaneously. This assessment and interview support department follows up by registering tasks in the task management database after the interview. This assessment and interview support department automatically sends reminder notifications on designated dates. This assessment and interview support department achieves automatic elimination of inappropriate input and streamlines interview processes through these processes. This assessment and interview support department powerfully supports regular one-on-one meetings through digital technology. This assessment and interview support department replaces processes relying on human memory and manual operations with reliable computer data management and automated notifications. This assessment and interview support department improves the quality and traceability of communication within the organization.

[0107] (specific use case) For example, imagine a nursing facility where multiple managers oversee their respective staff. The data collection department automatically retrieves personnel information for each manager, including their staff's hiring and leaving dates, transfer intentions, job transfer history, training history, qualification status, years of continuous service, past performance reviews, promotion history, and disciplinary records. Additionally, it retrieves detailed nursing information from the nursing record system, such as each employee's nursing plan achievement rate, daily nursing records, nursing content for each user, frequency of nursing interventions, and details of feeding / bathing / toilet assistance / rehabilitation support / recreation implementation. It also retrieves user satisfaction survey results, each user's nursing satisfaction score, and records of direct gratitude and complaints from users. Furthermore, it retrieves information from the incident / near-incident reporting system, including the number of incidents, date and time of occurrence, location, cause, response measures, recurrence prevention strategies, involved employees, incident severity classification, and follow-up status. As a record of communication among employees, this data collection department gathers information such as business contact systems and chat logs, meeting attendance records, business handover records, employee consultation / suggestion records, meeting minutes of regular meetings, and exchanges of thank-you / appreciation messages among employees. Furthermore, as a result of regular stress checks and health diagnoses, it also collects stress levels, health risk scores, guidance from doctors, implementation status of health maintenance activities, and mental health counseling records. As a result of anonymous feedback and 360-degree evaluations from users or their families, this department collects information from questionnaire systems and dedicated feedback systems, including opinions expressed freely, scores based on 5- or 10-level ratings, individual comments on specific employees, letters and emails from users' families, thank-you notes and complaints, feedback submission frequency, and time-series changes in feedback. Specifically, this data collection department performs parallel data acquisition and processing from multiple endpoints. It uses multi-threading to process data requests sent to different systems simultaneously. This department hides network latency and shortens the overall data acquisition time. This department applies data cleaning algorithms to the acquired data in various formats. This data collection department performs normalization processes such as date format standardization, conversion between full-width and half-width characters, and removal of unnecessary whitespace. This data collection department executes missing value imputation algorithms. This data collection department uses past data trends and data from similar employees to statistically estimate and imput missing values. This data collection department uses natural language processing to remove noise from free-flowing text. This data collection department performs stop word removal and stemming. This data collection department completely eliminates inconsistencies during data integration through these preprocessing steps. This data collection department ensures the quality of input data in subsequent AI analysis. This data collection department significantly improves the inference accuracy of AI models by providing high-quality data with less noise. This data collection department technically solves the data integration challenges in large-scale medical corporations and nursing operators managing multiple facilities with heterogeneous systems.This data collection department utilizes high-speed computer processing to instantly complete data shaping tasks that would otherwise take a long time to do manually. This data collection department achieves the clear technological benefit of improved efficiency and accuracy in data processing.

[0108] The Database Management Department organizes this data according to multiple entry points, such as each manager, employee, user, department, facility, and event, and chronologically. This allows for the extraction and aggregation of data such as: the turnover rate of each manager's subordinates over the past three years; the annual comparison of transfer intention rates for each facility; the time-series changes in the nursing quality scores of specific employees; the monthly trend of the number of events; the time-series changes in anonymous feedback; the correlation analysis between training experience and nursing quality scores; the relationship between the frequency of inter-employee communication and the event occurrence rate; the correlation between user satisfaction and nursing plan achievement rate; the relationship between employee health risk scores and turnover rate; and the relationship between the frequency of feedback from family members and the number of complaints. Specifically, the Database Management Department constructs a multidimensional online analysis and processing cube. The department pre-aggregates data along various dimensions such as managers, employees, facilities, and time. The pre-aggregated data is stored as a multidimensional array in memory. The response time for complex aggregation queries is reduced to milliseconds. A data caching mechanism is installed in the in-memory database. The results of frequently executed retrieval queries are maintained in memory. This database management system eliminates recalculation for the same query. It applies a dedicated compression algorithm to time-series data. It performs incremental encoding, recording only the changing portions of the data. It significantly reduces storage capacity consumption. It employs a distributed database architecture. Even with increasing data volume, it horizontally scales the system by adding storage nodes. These processes enable high-speed retrieval and aggregation of large and complex datasets. It eliminates waiting time for executives and managers performing ad-hoc data analysis. It provides a foundation for rapid, data-driven decision-making. It offers solutions through improvements in computer technology to address the complex multidimensional and time-series analysis requirements that existing purely relational databases struggle to handle.

[0109] The Generative AI Analytics Department uses this data to calculate the turnover rate and transfer intention rate of each manager's subordinates. For example, it outputs quantitative indicators such as "Among Manager A's 10 subordinates, 2 left in the past year, and 1 expressed a transfer intention, so the turnover rate is 20% and the transfer intention rate is 10%." Regarding nursing quality, it comprehensively evaluates nursing plan achievement rate, user satisfaction questionnaires, training experience, qualification status, diversity of nursing content, frequency of nursing implementation, nursing satisfaction score for each user, and records of user thanks / complaints, outputting multiple indicators such as "The average nursing quality score of subordinates is 4.3 points," "The nursing plan achievement rate is 95%," "The average user satisfaction score is 4.5 points," and "The diversity score of nursing content is 0.8." Regarding accidents / near misses, it calculates the number of reports, major incident incidence rate, and recurrence rate for each employee, deriving detailed indicators such as "The incident incidence rate is 0.2 incidents / person / year," "The major incident incidence rate is 0.05 incidents / person / year," and "The incident recurrence rate is 10%." Furthermore, regarding anonymous feedback from users or their families, generative AI uses natural language processing to analyze the text, extracting keywords and emotions such as "trustworthy," "friendly," "careful explanation," "slow response," "feeling uneasy," "dissatisfied with nighttime responses," "employee's smile is reassuring," "careful feeding assistance," and "insufficient care during elimination assistance." Trust scores and improvement point lists are generated for each manager. In cases with many negative comments, specific improvement proposals are automatically generated, such as "insufficient explanation from employees," "feeling uneasy about nighttime responses," "few conversations during feeding assistance," and "insufficient care during elimination assistance." The distribution of employee evaluation data and the tendency of each attribute are analyzed to detect bias risks such as "consistently giving low ratings to female employees," "giving high ratings to specific qualification holders," "strict ratings to newly hired employees," and "lenient ratings to long-term continuous employees," calculating bias scores. The relationship between employee health risk scores and stress test results and turnover and incident rates is also analyzed, extracting causal relationships such as "employees with high stress scores have a 1.5 times higher turnover rate than average" and "employees with high health risks have a 2 times higher incident rate." Specifically, this generative AI analysis unit executes a multimodal neural network and a gradient boosting decision tree. This generative AI analysis unit possesses an architecture that comprehensively processes both textual and numerical data. This generative AI analysis unit receives embedding vectors of text and normalized vectors of numerical data as input data. As a specific example of input data, this generative AI analysis unit receives a one-dimensional tensor representing a normalized value of 0.2 for the turnover rate, and a 768-dimensional embedding vector transformed from the text "slow response" using a natural language processing model. This generative AI analysis unit integrates these heterogeneous data into a single feature vector in a consolidation layer. This generative AI analysis unit then inputs the integrated feature vector into a fully connected layer.This generative AI analysis unit generates bias scores and causal relationship weighting coefficients as output data. The output data format includes a bias score represented by a floating-point number ranging from 0.0 to 1.0, and a weighting coefficient vector representing the contribution of each feature. As a specific example of the output data, this unit outputs a bias score of 0.8 and a weighting coefficient of 0.6 for the stress score. This unit forwards the output results to subsequent processing. If the bias score exceeds a threshold, the unit initiates a data correction module. This unit dynamically generates a causal relationship graph based on the weighting coefficients. During model learning, this unit minimizes the loss function that combines cross-entropy error and regularization terms. These processes ensure the fairness of the evaluation. This unit extracts hidden causal relationships that are difficult for humans to detect through computation in high-dimensional space. Through the unconventional method of automatically detecting evaluation bias, this unit achieves a significant improvement in the objectivity of personnel evaluations.

[0110] Based on these analysis results, the evaluation report generation department automatically generates an evaluation report containing specific values ​​and comments such as "Manager A's turnover rate is 20%, nursing quality score is 4.3, nursing plan achievement rate is 95%, incident rate is 0.2 incidents / person / year, major incident rate is 0.05 incidents / person / year, trust score is 0.75, improvement points: need to strengthen nighttime response, should increase conversations during feeding assistance, thoroughly implement care during elimination assistance, anonymous feedback: 'detailed explanation,' 'feeling uneasy about nighttime response,' 'detailed feeding assistance,' 'feeling uneasy during elimination assistance.'" The report is automatically updated monthly or quarterly and can be viewed by the manager, superiors, and operations headquarters. In cases of a sharp increase in turnover and incident rates, or a sharp drop in trust scores, the Alert Notification Department sends real-time warnings to relevant personnel, notifying them of specific situations and recommended actions, such as "The turnover rate of subordinate employees has exceeded 30% in the past three months. Urgent action is required," "We have received numerous comments in anonymous feedback expressing 'unease about nighttime response.' We recommend re-evaluating the nighttime system," and "The incident rate has increased by 2 times compared to last month. Please reassess employee health and workload." Specifically, the Evaluation Report Generation Department executes a dynamic content generation engine. This department reads various scores and text data output from the Generative AI Analysis Department as structured objects into memory. It parses these structured objects and constructs hypertext files according to a defined report pattern. The department works in conjunction with a real-time stream processing framework. Once an update to the data source is detected, the department immediately regenerates the corresponding report content. The Alert Notification Department operates on an event-driven architecture. It receives anomaly detection events via a message queue system. The department determines the notification route based on the event's importance level. For high-importance events, this alert notification department executes simultaneous instant push notifications and emergency emails. For low-importance events, these are summarized and reported in daily batch processing. The evaluation report generation department and the alert notification department minimize report generation latency through these processes. They optimize network bandwidth consumption. They consistently display up-to-date and important information without delay in a comprehensive dashboard for regional managers overseeing multiple facilities. Through computer technology, they ensure information freshness and reliable delivery, enhancing the organization's crisis management capabilities.

[0111] The User Interface (UI) department allows managers, employees, and operations headquarters managers to access the server via a browser or dedicated application to view evaluation reports, publish feedback, conduct evaluation interviews, and confirm improvement proposals. Managers can review their own evaluation reports, understand their trust scores and areas for improvement, and thus re-evaluate their management practices. Employees can anonymously publish feedback, review their past feedback history, and track their career growth. Operations headquarters and HR managers can assess evaluation biases and risks across the entire facility or multiple facilities and apply this information to system improvements and talent optimization. Specifically, this UI department uses a single-page application framework. It utilizes a virtual document object model to accelerate screen rendering. It employs a Representational State Transition (RST) API for communication with the server. It implements a secure authentication mechanism using JSON Web Tokens. The UI department encrypts and stores the tokens issued upon user login in local storage. It attaches the token to the request header each time a request is sent to the server. The UI department verifies the token signature on the server side to ensure it has not been tampered with. It references a role-based access control list. This user interface dynamically restructures the interface so that general employees can only view their own data, while managers can view the data of their subordinates. Through these processes, this user interface effectively prevents information leakage. This user interface establishes secure communication channels to ensure the security of data in the personnel evaluation system that processes personal information. This user interface automatically controls the appropriate disclosure of information commensurate with user permissions. This user interface reliably and quickly executes complex access control as an internal computer technology process.

[0112] Specific examples of prompt texts read by generative AI include: "Calculating the turnover rate, transfer intention rate, nursing quality score, and incident rate of manager A's subordinates over the past year; conducting sentiment analysis of anonymous feedback to extract trust scores and improvement points"; "Predicting the likelihood of improved nursing quality in the next six months based on employee B's nursing records, training history, and satisfaction questionnaire results"; "Analyzing manager C's employee evaluation data by attribute, detecting any evaluation bias, and calculating bias scores"; and "Analyzing anonymous feedback text data through natural language processing, summarizing negative opinions, and generating improvement proposals." Specifically, this generative AI analysis department executes a dynamic prompt generation algorithm based on prompt engineering technology. This generative AI analysis department reads predefined prompt templates from the database. This generative AI analysis department vectorizes the attribute information and past evaluation history of the target manager as context data. This generative AI analysis department embeds the context data into the variable part of the template. This generative AI analysis department sends the completed prompt string as input data to a large-scale language model. This generative AI analysis unit, as a specific example of input data, receives a string tag sequence: "Generating an improvement proposal based on Manager A's turnover rate of 0.2 and nursing quality score of 4.3." This generative AI analysis unit executes the inference engine of a large-scale language model. This generative AI analysis unit receives a text tag sequence generated as output data. This generative AI analysis unit, as a specific example of output data, outputs the text tag sequence "Recommend strengthening nighttime coping strategies." This generative AI analysis unit applies a hallucination detection filter to the output text. This generative AI analysis unit uses a logic checking algorithm to verify whether the generated text contradicts factual data. This generative AI analysis unit suppresses AI hallucination phenomena through these processes. This generative AI analysis unit significantly improves the stability and reliability of the output. This generative AI analysis unit safely and accurately performs automatic text generation based on diverse evaluation benchmarks. This generative AI analysis unit eliminates the need for manual creation of prompts by humans, achieving complete system-based automation.

[0113] Thus, by using the system of this invention, management evaluations of nursing care can be visualized comprehensively and objectively, strongly supporting employee retention, improved nursing quality, increased user satisfaction, and sound organizational operations. Specifically, this system implements an orchestration engine that comprehensively controls the data flow between modules. The system re-accumulates evaluation results and user feedback in the database management department. The system utilizes the accumulated data as relearning data for the generative AI analysis department. The system executes a continuous machine learning pipeline. The system fine-tunes the model's weight parameters each time new data is added. The system autonomously improves the accuracy of evaluations and predictions over time. The system forms a feedback loop. The system continuously monitors the performance of the entire system, automatically detects bottlenecks, and dynamically allocates resources. Through these processes, the system functions as a self-regulating optimization system that transcends a mere operational efficiency tool. This system makes it possible to build a standardized personnel evaluation platform throughout the nursing industry. This system directly contributes to solving societal challenges through improvements in computer technology. This system seamlessly integrates the entire process from data collection to analysis, report generation, and feedback, powerfully promoting the soundness of organizational operations through advanced information processing technology.

[0114] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0115] Data generation model 58 is what is known as generative AI (Artificial Intelligence).

[0116] One example of data generation model 58 includes ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> Generative AI, etc.

[0117] The data generation model 58 is obtained by performing deep learning on a neural network. A prompt containing instructions is input to the data generation model 58, along with inference data such as speech data representing speech, text data representing text, and image data representing images (e.g., data from still images or moving images). The data generation model 58 infers the input inference data based on the instructions shown in the prompt and outputs the inference result in one or more data forms, such as speech data, text data, and image data. The data generation model 58 may include, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization. The specific processing unit 290 performs the aforementioned specific processing while using the data generation model 58. The data generation model 58 may also be a fine-tuned model to output inference results from prompts without instructions; in this case, the data generation model 58 can output inference results from prompts without instructions. The data processing device 12, etc., includes multiple data generation models 58, including AI other than generative AI. AI other than generative AI includes, but is not limited to, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes. Furthermore, AI can also act as an AI agent. Additionally, when the processing described above is performed by AI, this processing may be partially or entirely performed by AI, but is not limited to this example. Moreover, processing performed by AI, including generative AI, can be replaced by rule-based processing, and vice versa.

[0118] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart device 14 or external devices, and the smart device 14 acquires or collects information required for processing from the data processing device 12 or external devices.

[0119] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart device 14.

[0120] [Second Implementation] Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0121] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. One example of the data processing device 12 is a server.

[0122] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0123] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, and communication I / F 44 are also connected to the bus 52.

[0124] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0125] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0126] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0127] Figure 4 An example of the main functions of the data processing device 12 and the smart glasses 214 is shown. Figure 4As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0128] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0129] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0130] In the smart glasses 214, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290.

[0131] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart glasses 214. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0132] (Example 1) Since the specific processing flow is the same as that in Embodiment 1 described in the first embodiment above, the description is omitted.

[0133] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0134] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0135] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or external devices, and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or external devices.

[0136] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart glasses 214.

[0137] [Third Implementation] Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0138] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. An example of the data processing device 12 is a server.

[0139] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0140] The head-mounted terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, display 343, and communication I / F 44 are also connected to the bus 52.

[0141] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0142] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0143] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0144] Figure 6 An example of the main functions of the data processing device 12 and the head-mounted terminal 314 is shown. Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0145] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0146] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the special processing unit 290.

[0147] In the head-mounted terminal 314, the processor 46 performs the receiving and output processing.

[0148] The memory 50 stores a receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0149] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the head-mounted terminal 314. In the following description, the data processing device 12 will be referred to as the "server" and the head-mounted terminal 314 will be referred to as the "terminal".

[0150] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0151] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0152] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0153] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the head-mounted terminal 314, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the head-mounted terminal 314. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the head-mounted terminal 314 or external devices, and the head-mounted terminal 314 acquires or collects information required for processing from the data processing device 12 or external devices.

[0154] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the head-mounted terminal 314.

[0155] [Fourth Implementation] Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0156] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[0157] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0158] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control object 443. Computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, control object 443, and communication I / F 44 are also connected to the bus 52.

[0159] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0160] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0161] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0162] The controlled object 443 includes a display device, LEDs for the eyes, and motors for driving the arms, hands, and feet. The posture and movements of the robot 414 are controlled by controlling the motors for the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, facial expressions of the robot 414 can also be expressed by controlling the illumination state of the LEDs for the robot 414's eyes.

[0163] Figure 8 An example of the main functions of the data processing device 12 and the robot 414 is shown. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0164] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0165] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the special processing unit 290.

[0166] In robot 414, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on RAM 48.

[0167] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the robot 414. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 will be referred to as the "terminal".

[0168] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0169] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0170] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0171] Furthermore, the processing performed by the data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the robot 414, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the robot 414. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or external devices, and the robot 414 acquires or collects information required for processing from the data processing device 12 or external devices.

[0172] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the robot 414.

[0173] Furthermore, the emotion-specific model 59, serving as an emotion engine, can determine the user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine the user's emotion based on an emotion graph that serves as a specific mapping (see...). Figure 9 The system determines the user's emotions. In addition, the emotion-specific model 59 can also determine the robot's emotions in the same way, and the specific processing unit 290 can also perform specific processing using the robot's emotions.

[0174] Figure 9 This is a diagram representing an emotion map 400 that maps various emotions. In the emotion map 400, emotions are arranged radially from the center in concentric circles. The closer to the center of the concentric circles, the more primitive the emotion is. Further out on the concentric circles, emotions are arranged representing states or actions arising from mood. Emotion is a concept that includes both feelings and mental states. To the left of the concentric circles, emotions generated by reactions occurring in the brain are arranged roughly. To the right of the concentric circles, emotions guided by situational judgments are arranged roughly. Above and below the concentric circles, emotions generated by reactions occurring in the brain and guided by situational judgments are arranged roughly. Furthermore, the emotion of "pleasure" is arranged above the concentric circles, and the emotion of "unpleasantness" is arranged below. Thus, in the emotion map 400, various emotions are mapped according to the structure of emotion generation, while easily generated emotions are mapped nearby.

[0175] These emotions are distributed at the 3 o'clock position on the Emotion Chart 400, and usually fluctuate between peace and unease. In the right half of the Emotion Chart 400, because situational awareness is more dominant than internal feelings, it gives a sense of calm.

[0176] The inner side of the emotion diagram 400 represents the mind, and the outer side of the emotion diagram 400 represents actions. Therefore, the further you go to the outer side of the emotion diagram 400, the more the emotion can be seen (manifested in actions).

[0177] Here, human emotions are based on a balance of various factors such as posture and blood sugar levels. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. In robots, cars, and motorcycles, emotions can also be created based on a balance of factors such as posture and remaining battery power. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Speech Emotion Recognition and Brain Physiological Signal Analysis Systems for Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the "Reaction" domain, where sensation is dominant, are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the "Situation" domain, where situational cognition is dominant, are arranged.

[0178] The emotion map defines two types of emotions that promote learning. One is a negative emotion located near the middle of "repentance" or "reflection" on the situation side. That is, when the robot experiences negative emotions such as "I never want to feel this way again" or "I never want to be scolded again." The other is a positive emotion located near "desire" on the response side. That is, when the robot experiences positive feelings such as "wanting more" or "wanting to know more."

[0179] The emotion-specific model 59 feeds user input into a pre-learned neural network to obtain emotion values ​​representing each emotion shown in the emotion graph 400, and determines the user's emotion. This neural network is pre-learned based on multiple learning data sets that combine user input with emotion values ​​representing each emotion shown in the emotion graph 400. Furthermore, this neural network is learned to... Figure 10 As shown in sentiment graph 900, nearby sentiment values ​​are similar to each other. Figure 10 Examples show how emotions such as "peace of mind," "stability," and "reassurance" can result in similar emotional values.

[0180] The above description primarily focuses on the functions of the data processing device 12 to illustrate the system disclosed herein. However, the system disclosed herein may not necessarily be implemented on a server. The system disclosed herein can be implemented as a general information processing system. For example, it can be implemented as a software program running on a personal computer, an application running on a smartphone, etc. The methods disclosed herein can be provided to users in the form of SaaS (Software as a Service).

[0181] In the above embodiments, an example of specific processing performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and distributed processing of specific processes by multiple computers, including computer 22, is also possible. For example, the data generation model 58 may also be provided in an external device of the data processing apparatus 12, where data is generated based on input data.

[0182] In the above embodiments, an example of storing a specific processing program 56 in memory 32 is illustrated, but the technology disclosed herein is not limited thereto. For example, the specific processing program 56 may also be stored in a portable computer-readable non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 performs specific processing according to the specific processing program 56.

[0183] Alternatively, the specific processing program 56 can be stored in a storage device such as a server connected to the data processing device 12 via a network 54, and the specific processing program 56 can be downloaded and installed into the computer 22 upon request from the data processing device 12.

[0184] Furthermore, it is not necessary to store the entire specific process 56 in a storage device such as a server connected to the data processing device 12 via the network 54, nor is it necessary to store the entire specific process 56 in the memory 32; a portion of the specific process 56 may also be stored.

[0185] As a hardware resource for performing specific processing, various processors can be used. For example, a CPU can be listed as a processor; it is a general-purpose processor that functions as a hardware resource for performing specific processing by executing software, i.e., a program. Furthermore, as a processor, a dedicated circuit can be listed as a processor; it is a processor with a circuit structure specifically designed for performing specific processing, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit).

[0186] Every processor has built-in or connected memory, and every processor performs specific processes by using memory.

[0187] The hardware resources for performing a specific process can consist of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware resources for performing a specific process can also be a single processor.

[0188] As an example of a single processor, the first type consists of a combination of one or more CPUs and software, which functions as a hardware resource to perform specific processing. The second type uses a processor, such as a System-on-a-chip (SoC), which implements the entire system functionality, including multiple hardware resources performing specific processing, using a single IC chip. In this case, the specific processing is implemented using one or more of the aforementioned processors that serve as hardware resources.

[0189] Furthermore, as the hardware architecture of these various processors, more specifically, circuits combining semiconductor elements and other circuit components can be used. Moreover, the specific process described above is merely an example. Therefore, it goes without saying that, without departing from the main point, unnecessary steps can be removed, new steps can be added, or the processing order can be changed.

[0190] The foregoing descriptions and illustrations are detailed explanations of the parts covered by this disclosure and are merely one example of this disclosure. For instance, the descriptions of the above-described structure, function, role, and effect are just one example of the structure, function, role, and effect of the parts covered by this disclosure. Therefore, it goes without saying that, without departing from the spirit of this disclosure, unnecessary parts can be deleted, new elements can be added, or replacements can be made to the foregoing descriptions and illustrations. Furthermore, to avoid confusion and facilitate understanding of the parts covered by this disclosure, explanations of technical common sense that does not require special explanation for implementing this disclosure have been omitted from the foregoing descriptions and illustrations.

[0191] All documents, patent applications and technical standards described in this specification are incorporated herein by reference as if they were specifically and individually described as incorporated by reference.

[0192] In addition, the following notes further explain the above explanation.

[0193] (Note 1) A nursing facility management evaluation system comprises a data collection unit, a generative AI analysis unit, and an evaluation report generation unit. The data collection unit automatically and comprehensively collects various information from multiple databases and systems both inside and outside the nursing facility, including personnel data, resignation / transfer intentions / job change history, training history, qualification status, nursing plan achievement rate, user satisfaction questionnaires, accident / near-accident reports, employee communication history, anonymous feedback from users or family members, and results of health diagnoses or stress tests. The generative AI analysis unit uses the data acquired by the data collection unit to quantitatively calculate each... The evaluation system collects data on employee turnover rate, transfer intention rate, care quality score, and incident rate for each manager's subordinates. It also uses natural language processing technology to analyze anonymous feedback from users or family members to extract trust scores and a list of improvement points. Furthermore, it analyzes the distribution of employee evaluation data and the tendency of each attribute to detect any subjective bias. The evaluation report generation department automatically generates an evaluation report for each manager, including the turnover rate, transfer intention rate, care quality score, incident rate, trust score, list of improvement points, and bias detection results obtained by the generative AI analysis department. This report is then distributed periodically or in real-time to the manager, their superiors, and the operations headquarters.

[0194] (Note 2) As described in Appendix 1, the generative AI analysis unit analyzes feedback data anonymously collected from users or their families using natural language processing technology, calculates trust scores, automatically summarizes negative opinions and improvement requests, extracts the current status of trust relationships for each manager and specific areas for improvement, and provides these to the evaluation report generation unit to visualize and propose improvement measures for the trust relationships between managers, their subordinates, and users.

[0195] (Note 3) As described in Appendix 1, the system wherein the evaluation report generation unit automatically generates evaluation reports on a regular or real-time basis based on the analysis results of the generative AI analysis unit. These reports include each manager's turnover rate, transfer willingness rate, care quality score, incident rate, bias detection results, trust score, and list of improvement points. The reports are distributed in a format that can be viewed by the manager, their superiors, and the operations headquarters. The transparency of the evaluation process is ensured so that the reports can be used as an objective basis for managers and their subordinates to conduct evaluation interviews.

Claims

1. A personnel evaluation system, comprising a data collection unit, a generative AI analysis unit, and an evaluation report generation unit, characterized in that, The data collection department automatically collects personnel-related data from multiple databases both inside and outside the organization, including employee turnover, job hopping, transfer intentions, promotion performance, performance evaluations, potential scores, and anonymous feedback for each manager. The generative AI analysis department uses the data collected by the data collection department to quantitatively calculate the turnover rate, transfer intention rate, and promotion rate of each manager's subordinates. Based on the performance and potential of subordinates, it predicts future promotion possibilities and further analyzes the distribution of evaluation scores given by managers and the evaluation bias of each attribute to detect any subjective bias. Additionally, it extracts managers' management style, trust scores, and areas for improvement through natural language analysis of anonymous feedback. The evaluation report generation department automatically generates evaluation reports for each manager, including turnover rate, transfer intention rate, promotion rate, bias detection results, trust scores, and a list of improvement points obtained by the generative AI analysis department, and distributes these reports to managers and management departments periodically or in real-time.

2. The system as described in claim 1, characterized in that, The generative AI analysis department uses natural language processing technology to analyze feedback data anonymously collected from subordinates, calculates trust scores, and automatically summarizes negative opinions and improvement requests. It extracts the current status of trust relationships for each manager and specific areas for improvement. By providing these to the evaluation report generation department, it enables visualization of the trust relationship between managers and subordinates and generates improvement proposals.

3. The system as described in claim 1, characterized in that, Based on the analysis results of the generative AI analysis department, the evaluation report generation department regularly or in real time automatically generates evaluation reports containing information such as turnover rate, transfer intention rate, promotion rate, bias detection results, trust score, and improvement point list for each manager. The reports are distributed in a format that can be viewed by the managers themselves, their superiors, and the human resources department, and the transparency of the evaluation process is ensured so that they can be used as an objective basis for managers and their subordinates to conduct evaluation interviews.

Images