Computing systems and methods of rapidly training and executing artificial intelligence agents

Abstract

Systems and methods are provided for rapidly training an Al model. An industry application Al model is generated by retraining a fundamental Al model using industry¬ specific data. A client Al model is generated by retraining the industry application Al model using client-specific data, including control data. The client Al model is executed by obtaining data from one or more data nodes in a client ecosystem process, which generates a score, or an analysis, or a recommendation, or a combination thereof.

Description

Atty. Dkt. No. 142937-5002COMPUTING SYSTEMS AND METHODS OF RAPIDLY TRAINING AND EXECUTING ARTIFICIAL INTELLIGENCE AGENTSCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This patent application claims priority to United States Provisional Patent Application No. 63 / 718,537, filed on November 8, 2024, and titled “COMPUTING SYSTEMS AND METHODS FOR RAPIDLY TRAINING AND EXECUTING ARTIFICIAL INTELLIGENCE AGENTS”, the entire contents of which are herein incorporated by reference.TECHNICAL FIELD

[0002] The disclosed exemplary embodiments relate to computer-implemented systems and methods for rapidly training and executing artificial intelligence agents.BACKGROUND

[0003] In many cases, training artificial intelligence (Al) models to conduct specific analysis or to execute certain logic uses a large amount of computing resources (e.g., processor and memory resources). In many cases, the training process also requires personnel with skills to develop a training model pipeline, and to deploy the trained Al model. In some cases, data pipelines include creating data connectors and APIs between data sources, which can create computing complexity. The process for establishing an Al model and training the same also takes a significant amount of time.

[0004] In some cases, training an Al model also uses large amounts of data. In some cases, users do not have access to a sufficient amount of data or the appropriate type of data to develop an effective Al model.

[0005] In some cases, the data is too large and consequently causes various forms of hallucinations.

[0006] In some cases, the user does not know how to write an effective prompt for a large language model (LLM), leading to inaccurate responses.

[0007] In some cases, the user does not how to import, incorporate and generate analysis, outcomes, and recommendations with explainability using structured data, unstructured data, third party data, and Internet data. In some cases, existing computing systems involve many separate computing software modules andAtty. Dkt. No. 142937-5002 many data sources, which have their own user interfaces, but do not communicate with each other.

[0008] In some cases, a client has their own data that is considered proprietary, confidential or private. Using this client data to train an Al model may create risks for the private data to be released to third parties. In some cases, executing the client data is limited and is not able to be used to train an effective Al model.SUMMARY

[0009] The following summary is intended to introduce the reader to various aspects of the detailed description, but not to define or delimit any invention.

[0010] In at least one broad aspect, there is provided a computing system for rapidly training and executing an Al agent.

[0011] In at least another broad aspect, a method is provided for rapidly training and deploying an Al agent, the method executed in a computing environment comprising a processor, a memory and a communication interface.

[0012] In at least another broad aspect, a computing system is provided for training and deploying artificial intelligence (Al) models, the computing system comprising a processor, a communication interface, and memory, the processor coupled to the communication interface and the memory. The memory comprises a fundamental Al model that is trained. The processor configured to at least: ingest, from an external data source, industry-specific data; re-train a fundamental Al model using the industry-specific data to generate an industry-specific Al model; deploy the industry-specific Al model; ingest, from a client data source, client data; re-train the industry-specific Al model using the client data to generate a client Al model; deploy the client Al model; ingest, from the client data source, new client data; and, re-train the client Al model using the new client data to generate a new client Al model, and deploy the new client Al model in place of the client Al model.

[0013] In some cases, the computing system further comprises a client Al computing environment and an Al computing environment separate from the client Al computing environment; wherein the industry-specific Al model is deployed within the client Al computing environment; wherein the client Al model is generated and deployed within the client Al computing environment; wherein the new client Al modelAtty. Dkt. No. 142937-5002 is generated and deployed within the client Al computing environment; and wherein the industry-specific Al model is generated in the Al computing environment.

[0014] In some cases, the client Al computing environment comprises one or more security permissions to correspond with the client data source; and wherein and the one or more security permissions, the client data source, the client Al model, and the new client Al model are inaccessible by the client Al computing environment.

[0015] In some cases, the computing system further comprises a second client Al computing environment that is separate from the client Al computing environment and the Al computing environment. The processor is further configured to: deploy the industry-specific Al model in the second client Al computing environment; ingest, from a second client data source accessible by the second client Al computing environment, second client data; re-train the industry-specific Al model using the second client data to generate a second client Al model; deploy the second client Al model in the second client Al computing environment; ingest, form the second client data source, new second client data; re-train the second client Al model using the new second client data to generate a new second client Al model, and deploy the new second client Al model in place of the second client Al model in the second client Al computing environment.

[0016] In some cases, the client data source is part of a client system, and the client Al computing environment comprises a digital container that establishes secure communication with the client system.

[0017] In some cases, the processor is further configured to: after deploying the new client Al model, ingest, from the external data source, new industry-specific data; re-train the fundamental Al model or the industry-specific Al model, or both, using the using the new industry-specific data to generate a new industry-specific Al model; deploy the new industry-specific Al model; re-train the new industry-specific Al model using the new client data to generate a second new client Al model; and deploy the second new client Al model in place of the new client Al model.

[0018] In some cases, ingesting the client data comprises: the computing system communicating with multiple data nodes of the client data source; wherein the multiple data nodes comprises: a first data node that generates a first output; a second data node that is configured to process at least the first output to generate a secondAtty. Dkt. No. 142937-5002 output; and a third data node that is configured to process at least the second output to generate a third output; and, wherein the first output, the second output and the third output form at least part of the client data.

[0019] In some cases, ingesting the client data comprises: the computing system communicating with multiple data nodes of the client data source; and wherein one or more of the multiple data nodes comprises a node Al agent that is configured to generate node data that forms at least part of the client data.

[0020] In some cases, the multiple data nodes comprise multiple robots.

[0021] In some cases, the multiple data nodes comprise multiple vehicles.

[0022] In some cases, the multiple data nodes comprise multiple machines.

[0023] In some cases, the new client Al model communicates with a client computing device, and, when the new client Al model is deployed, the new client Al model is configured to generate a message that is transmitted to the client computing device; and, wherein the message comprises an analysis of at least the client data and the new client data.

[0024] In some cases, the message is automatically generated in response to the computing system detecting a query from the client computing system.

[0025] In some cases, the message is automatically generated in response to the computing system detecting new industry specific data that differs from the industry-specific data.

[0026] In some cases, the message is automatically generated in response to the computing system detecting additional new client data that differs from the new client data and the client data.

[0027] In some cases, the processor is further configured to at least: ingest control data comprising natural language; process the control data using the client Al model or the new client Al model to generate a plurality of natural language logic conditions; when the client Al model or the new client Al model is deployed, receive the client data from a data node that is part of the client data source; determine, using the client Al model, a node data score relevant to one or more of the natural language logic conditions; and, determine if the node data scoreAtty. Dkt. No. 142937-5002 triggers an urgent condition, and, responsive to triggering the urgent condition, generating and sending a message comprising information about the data node.

[0028] In some cases, the message comprises a natural language description regarding the data node, contextual information obtained from the control data, and compliance or non-compliance based on the one or more of the natural language logic conditions.

[0029] In some cases, when ingesting the client data from the client data source, the processor is configured to provide a graphical user interface comprising a drag-and-drop data collection object; wherein the drag-and-drop data collection object is configured to receive one or more data files via a drag-and-drop interaction, the one or more data files comprising the client data.

[0030] In at least another broad aspect, a method for training and deploying artificial intelligence (Al) models is provided. The method is executed in a computing environment comprising a processor, a communication interface, and memory, the processor coupled to the communication interface and the memory. The method comprises: storing a fundamental Al model that is trained in the memory; ingesting, from an external data source, industry-specific data; re-training the fundamental Al model using the industry-specific data to generate an industry-specific Al model; deploying the industry-specific Al model; ingesting, from a client data source, client data; re-training the industry-specific Al model using the client data to generate a client Al model; deploying the client Al model; ingesting, from the client data source, new client data; and, re-training the client Al model using the new client data to generate a new client Al model, and deploying the new client Al model in place of the client Al model.

[0031] According to some aspects, the present disclosure provides a non- transitory computer-readable medium storing computer-executable instructions. The computer-executable instructions, when executed, configure a processor to perform any of the methods described herein.Atty. Dkt. No. 142937-5002BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings included herewith are for illustrating various examples of articles, methods, and systems of the present specification and are not intended to limit the scope of what is taught in any way. In the drawings:FIG. 1 is a schematic block diagram of a computing system for rapidly training Al models in accordance with at least some embodiments;FIG. 2A is a schematic block diagram of a computing system for processing data from one or more data nodes using a client Al agent, and providing outputs to a client user computing device in accordance with at least some embodiments;FIG. 2B is a schematic block diagram of another computing system for processing data from one or more data nodes using a client Al agent, and providing outputs to a client user computing device in accordance with at least some embodiments;FIG. 3 is a schematic block diagram of a computing system for training a client Al agent, and executing the same in accordance with at least some embodiments;FIG. 4 is a schematic block diagram of a data and computing architecture for training client Al agents, and executing the same in accordance with at least some embodiments;FIG. 5 is a flowchart diagram of an example method of training a client Al model, in accordance with at least some embodiments;FIG. 6 is a flowchart diagram of an example method of executing a client Al model, in accordance with at least some embodiments;FIG. 7 is a graphical user interface (GUI) for uploading control data files to train a client Al model, in accordance with at least some embodiments;FIG. 8 is a GUI displayable at a data node for uploading data files into a trained client Al model for processing, in accordance with at least some embodiments;FIG. 9 is a schematic block diagram of a system of devices for ingesting data by a data node, in accordance with at least some embodiments; andFIG. 10 is a schematic block diagram of a computer, in accordance with at least some embodiments.Atty. Dkt. No. 142937-5002DETAILED DESCRIPTION

[0033] Referring to FIG. 1 , a computing system 100 is hearing provided for rapidly training Al models and executing the same. An Al computing environment 110 communicates with one or more external data sources 150 to ingest various types of data.

[0034] In some cases, data from the external sources is used to train a fundamental Al model 112. In some cases, industry specific data is used to train one or more industry application Al models. For example, a trained industry application A (or industry specific) Al model 114 is relevant to a particular industry A, and a trained industry application B (or industry specific) Al model 114 is relevant to a particular industry B. For example the industries A and B are different industries. Some examples of different industries include: supply chain industry, manufacturing industry, oil and gas industry, logging industry, data security industry, etc. In some cases, an Al model library stores multiple trained Al models that trained to be specific to different industries and / or industry applications.

[0035] In some cases, data ingested from the external data sources is converted by an Al model (which may be the fundamental Al model or an indexing Al model) to vector data. The vector data is stored in a vector database in the Al computing environment 110, as better shown in FIGs. 3 and 4. Storing vector data in a vector database reduces the amount of data that is stored. In some cases, the storing vector data in vector database reduces the number of computation steps for the Al model when retrieving the vector data and processing the same by a fundamental Al model and / or an industry application Al model.

[0036] In the example in FIG. 1 , a client Al computing environment 118 is provided that is dedicated to a given (e.g., Client 1 ). In some cases, the client Al computing environment 118 resides in a cloud computing environment shared with the fundamental Al model, and is external to the client-owned computing infrastructure. In some cases, for this configuration, a digital container in the Al computing environment 110 is used to secure and separate the client Al computing environment 118 of Client 1 from other client Al computing environments of other clients (e.g., Client 2, Client 3, etc.). In some cases, the digital container of the client Al computing environment isAtty. Dkt. No. 142937-5002 secured by one or more encryption keys, wherein the corresponding client system stores a private encryption key configured to generate a signature that is verifiable by the digital container using a corresponding public encryption key. In some cases, one or more passwords are used to permit access to the digital container by the corresponding client system.

[0037] In some cases, the client Al computing environment 118 resides within the client-owned computing infrastructure. This is also called on-premise (or on-prem) computing. For example, the client owns their own private cloud computing environment (e.g., made available by Azure, Amazon Web Services, Google, etc.), and the client Al computing environment 118 resides in Client 1’s private cloud computing environment. This configuration may be desirable when the user wishes to maintain more security and privacy of their data (e.g., called client data).

[0038] In some other cases, the client Al computing environment 118 resides in a hybrid configuration, in which some of the computations are executed on the Al computing environment 118 are executed in the Al computing environment shared by clients and on Client 1 ’s private cloud computing environment.

[0039] The Al computing environment 118 for Client 1 includes a Client 1 Al model 120, and one or more Client 1 Al Agents 122 that execute the Client 1 Al model 120.

[0040] In some cases, Client 1 systems 170 interact with the Client 1 Al agents 122 and the Client 1 Al model 120. The Client 1 systems 170 include one or more Client 1 data sources 172 and one or more Client 1 user computing devices 174. In some cases, these user computing devices include on-prem devices and off-prem devices. Examples of user computing devices include desktop computers, tablets, laptops, smartphones, computing kiosks, computers within a vehicle, wearable computers (e.g., computing headsets, goggles, smart headphones, and smart watches), etc.

[0041] In operation, the fundamental Al model 112 is trained using data and analytics from many different data sources, to build a general capability and understanding.

[0042] One or more industry application Al models 114, 116 are trained to conduct analysis and execute logical conditions based on industry specific data. ThisAtty. Dkt. No. 142937-5002 creates a more specialized Al model for a given industry. In some cases, the industry application Al models are trained using external data sources with data specific to a given industry.

[0043] A client within the same given industry retrains the industry application A trained Al model 114 using their own data sources (e.g., Client 1 data sources 172) to more quickly train and refine a client specific Al model (e.g., Client 1 Al model 120). In this way, Client 1 ’s Al model 120 is quickly trained. From the perspective of Client 1 , only Client 1 ’s data is used to train the Client 1 Al model 120.

[0044] In some cases, the data from the Client 1 data sources 172 is indexed and stored as vectors in a vector database within the Al computing environment 118 for Client 1. The vectors data is not human readable and, therefore, improves ability to maintain privacy and security of the data belonging to Client 1 . Storing vector data in a vector database reduces the amount of data that is stored in association with Client 1 ’s computing resources. In some cases, the storing vector data in vector database reduces the number of computation steps for when the vector data is to be processed by the Client 1 Al model 120.

[0045] In some cases, the client specific Al model (e.g., Client 1 Al model 120) is also trained using third party data sources, which may be public, and internal data sources that are specific to Client 1. The data could describe and / or originate from upstream data nodes or downstream data nodes (or both) associated with the given client (e.g., Client 1 ). For example, an upstream data node could be operated by a vendor or supplier of the given client, and provide data describing operations of the vendor or supplier. For example, a downstream data node could be operated by a customer or distributor of the given client, and provide data describing operations of the customer or distributor.

[0046] In the example in FIG. 1 , a Client 2 Al model 126 is also trained using the same industry application A trained Al model 114. For example, Client 1 and Client 2 are both part of the same industry. The Client 2 Al model 126 and one or more Client 2 Al Agents 128 reside in Client 2’s Al computing environment 124, which communicates with Client 2 systems 180 (e.g., Client 2 data sources, and Client 2 user computing devices).Atty. Dkt. No. 142937-5002

[0047] A Client 3 Al model is trained based off the industry application B trained Al model 116, and communicates with Client 3 systems (e.g., Client 3 data sources, and Client 3 user computing devices).

[0048] In some cases, the client Al model is trained using control data that comprises unstructured data. In some cases, the control data comprises natural language data and other types of unstructured data, such as found in word documents, PDFs, videos, audio data, images, presentation slides, etc. In some cases, the control data includes contracts, policies, methodologies of thinking, guidelines, standard operating procedures, rules, statements of work, etc. During the training process, the control data is indexed and stored as vectors in a vector database. In some cases, the client Al model extracts natural language logic conditions from the control data and stores the same, or embeds the same within the Al model. In some cases, the natural language logic conditions are natural language prompts specific to the industry of the client and specific to the control data provided.

[0049] In some cases, the computing system combines data (e.g., data, libraries, model artifacts, or models, or a combination thereof) from the different industrial application trained Al models 114 and 116 to provide faster and more robust training of the client Al models or client Al agents, or both. In some cases, the computing system combines key performance indicators (KPIs), metrics, policy documents, standard operating procedure documents, customer data, structured data, unstructured data, and external 3rd party data (e.g., weather data, local / regional / country events, etc.) to accelerate training speed of a client Al model or a client Al agent, or both.

[0050] In some cases, the computing system automatically processes the data to generate anonymized data and composite indexed data, resulting in KPIs and metrics. In some cases, the computing system additionally generates and outputs the natural language text explaining the KPIs and metrics , and the natural language text includes proposed guardrails, potential or predicted outcomes, insights, and recommendations for a given client system 170.

[0051] After the trained client Al models are generated, they are deployed into a live production environment and run or executed using live data, in some cases, provided by client data sources. In some cases, the client data sources includeAtty. Dkt. No. 142937-5002 multiple data nodes. In some cases, deployment includes packaging a trained Al model into a format, like a container or a microservice, that integrates with the live production environment. In some cases, deployment of an Al model includes establishing data connections with existing software modules or computing devices, or both.

[0052] For example, in a supply chain industry, there a first data node represents a manufacturer of a product; a second data node represents a packing compony for packing the product; a third data node represents a shipping company for shipping the packaged product; a fourth data node represents a warehousing and distribution company for temporarily storing the packaged product before distributing the packaged product to stores; and multiple instances of a fifth data node represents multiple stores that receive the packaged product and sell the packaged product. These data nodes are within an process ecosystem of a given client. Data from each of the data nodes can provided to the client Al model to determine scoring according to the natural language logic conditions. For example, the client Al model can determine whether the different vendors of the different data nodes are compliant or non-compliant with any contracts, policies, methodologies of thinking, guidelines, standard operating procedures, rules, statements of work, etc. which form the control data.

[0053] In another example, a first data node represents a manufacturer of a product; upstream from this manufacturer there can be N number of data nodes such as raw material suppliers and subcontractor manufacturers that make components, which are in turn used by the final manufacture at the aforementioned first data node. A second data node represents N number of nationwide wholesalers or 3PL logistic warehouse firms that stock inventory for subsequent channel distribution to (a) smaller regional / local distributors; a third data node is the actual store(s) that sell the product to the businesses and customers. Between each of these nodes are N number of transportation data nodes representing the shippers moving product from raw materials and subcontractor component manufacturers to the primary manufacturer to the final manufacturer to downstream wholesalers, 3PLs, and ultimately the buyers. The ultimate buyers could have further downstream warehouse, logistics, and transportation data node. These aforementioned data nodes simplistically describe and define the technology ecosystems for a given business involved in a supply chain.Atty. Dkt. No. 142937-5002Depending on where a business sits in the aforementioned ecosystem affects the importance of certain data nodes. Data from each of the data nodes can provided to the client Al model to determine scoring according to the natural language logic conditions. For example, a client Al model can determine whether the different vendors of the different data nodes is compliant or non-compliant with any contracts, policies, methodologies of thinking, customer and vendor contracts, internal business guidelines, government I legal compliance, standard operating procedures, business rules, statements of work, etc. which form the control data.

[0054] This also applies to other industries, such as manufacturing. For example, the data nodes represent different manufacturing stations within a manufacturing process, and each manufacturing station produces data.

[0055] In some cases, the Al computing environment 110 includes: a tiered Al stack that includes the one or more fundamental Al models 112, which operate as a base tier; the one or more industry application trained Al models 114, 116, which operate as a mid-level tier with knowledge applicable to an industry; and, the one or more client Al models 120, 126 and the respective client Al agents 122, 128, which operate as a top tier with knowledge specific to a client in the industry.

[0056] In some cases, fundamental outputs from the one or more fundamental models 112 in the base tier are transmitted to the one or more industry application trained Al models 114, 116 in the mid-level tier, and the one or more industry application trained Al models 114, 116 process the fundamental outputs to generate industry specific outputs. The industry specific outputs from the one or more industry application trained Al models 114, 116 are transmitted to the one or more client Al models 120, 126 and the respective client Al agents 122, 128 in the top tier. The one or more client Al models 120, 126 and the respective client Al agents 122, 128 process the industry specific outputs to generate client specific outputs, which are transmitted to the one or more respective client systems. In some cases, the fundamental outputs from the one or more fundamental outputs are generated by processing a prompt that includes: data obtained from a given client system (e.g., 170, 180, 190), or data obtained from an external data source 150, or a first set of data from a given client system and a second set of data from an external data source.Atty. Dkt. No. 142937-5002

[0057] In some cases, client data from a given client system (e.g., 170) is transmitted to, and processed by, a respective client Al agent 122 and client Al model 120. The client Al agent 122 and client Al model 120 generate an industry specific prompt or query based on the client data, and the industry specific prompt or query is transmitted to a respective industry application trained Al model 114. The industry specific prompt, in some cases, includes industry data. The industry application trained Al model 114 uses the industry specific prompt or query to generate a generalized prompt, and transmits the same to the fundamental Al model 112. In some cases, the generalized prompt includes data from an external data source 150 or industry data, or both. The fundamental model 112 processes the generalized prompt and, in response, generates and outputs a fundamental output. The industry application trained model processes the fundamental output (from the fundamental output) and the industry specific prompt or query (from the client Al agent 122 and client Al model 120) to generate an industry specific output. In some cases, the process of generating the industry specific output includes processing industry specific data along with the fundamental output and the industry specific prompt. The industry specific output is transmitted to the client Al agent and the client Al model and, in response, the client Al agent and the client Al model processes the industry specific output to generate a client specific response. In some cases, the client Al agent and the client Al model process the client data along with the industry specific output to generate the client specific response. The client specific response is then returned to the given client system.

[0058] In some cases, the one or more fundamental Al models 112 in the base tier include large language models (LLMs). Some examples of these LLMs include those called by the trade names ChatGPT, Gemini, and Co-Pilot. In some cases, the one or more industry application trained Al models 114, 116 include: machine learning trained on specific industry knowledge; fine-tuned Al models specific to a given industry; Retrieval Augmented Generation (RAG) models with one or more libraries specific to a given industry; predefined prompts specific to a given industry; libraries of data specific to a given industry; Recurrent Neural Networks (RNNs) trained on data specific to a given industry; Generative Adversarial Networks (GANs) with a generator model or a discriminator model (or both) trained with data specific to a given industry;Atty. Dkt. No. 142937-5002 or Surface-Trend-Recommend-Infer-Predict-Action (STRIPA) computations; or one or more various combinations thereof.

[0059] Turning to FIG. 2A, the system 200 shows multiple data nodes 204A, 204B, 204B that are part of the Client 1 data sources 172. These data nodes are, in some cases, part of Client 1 ’s ecosystem process. In some cases, each data node transmits data to the Client 1 Al agent 122, which is processed by the Client 1 Al model 120, to determine a score and analysis relative to the control data. The components, including Client 1 Al model 120 and the Client 1 Al agent 122 that are in the client Al computing environment 118, are part of the Al computing environment 110 and the system 100 shown in FIG. 1.

[0060] In some cases, the output of a data node 204A is transmitted to a data node 204B for processing; and the output of the data node 20B is transmitted to the data node 204C for processing.

[0061] In some cases, the data provided by the data nodes includes one or a combination of: documents, video data, image data, audio data (e.g., speech data, or environmental audio data, or music, etc.), natural language data, structured data, machine data, and platform data (e.g., for Enterprise Resource Planning (ERP) platform, analytics platform, etc.). In some cases, different data nodes provide different types of data.

[0062] In some cases, the Client 1 Al model 120 is executed by the Client 1 Al Agent 122 to process the data from the data nodes, and to output a response to a query, to output an alert, or to output a report, or to output platform data, or a combination thereof.

[0063] In some cases, the process of uploading the data at a given data node is very easy. For example, using a GUI in a web browser application, or using an application, a user can drag and drop data files (e.g., photos, audio recordings, video, PDFs, reports, spreadsheets, custom data files, etc.) into the GUI for automatic processing by the Client 1 Al agent 122. The Client 1 Al agent 122 preprocesses the data (e.g., using optical character recognition, or speech-to-text processing) to extract the natural language data. In some cases, the Client 1 Al agent 122 preprocesses the data (e.g., image recognition processing, audio recognition processing, etc.) to extract relevant data for the Client 1 Al model. For example, video data of a conveyor beltAtty. Dkt. No. 142937-5002 shows product being moved. In other words, data can be used to train and / or execute a client Al model without application programming interfaces (APIs) or established data connectors.

[0064] Turning to FIG. 2B, a system 200’ that is similar to the system 200 is shown. The components, including Client 1 Al model 120 and the Client 1 Al agent 122 that are in the client Al computing environment 118, are part of the Al computing environment 110 and the system 100 shown in FIG. 1. In some cases, each of the data nodes 204A, 204B, 204C respectively include their own node Al agent(s) 205A, 205b, 205c. The node Al agents locally process data on the data node, using edge computing. In some cases, the node Al agents include and run small language models (SLMs). In some cases, the node Al agents communicate with the client Al agents or the client Al model in the client-dedicated Al computing environment 118.

[0065] In some cases, the data nodes are robotic devices configured to communicate with the client-dedicated Al computing environment 118 and, in some cases, also store and run a node Al agent. In some cases, the robotic devices are humanoid robots.

[0066] In some cases, the data nodes are machines configured to communicate with the client-dedicated Al computing environment 118 and, in some cases, also store and run a node Al agent. In some cases, the machines are computer numerically controlled machines. In some cases, the machines are 3D printers. In some cases, the machines include a series or network of material handler machines (e.g., pick and grab machines, conveyor machines, automated guided vehicles, autonomous mobile robots, stacker machines, etc.).

[0067] In some cases, the data nodes are vehicles configured to communicate with the client-dedicated Al computing environment 118 and, in some cases, also store and run a node Al agent. In some cases, the vehicles are autonomous vehicles. In some cases, the vehicles are unmanned vehicles. In some cases, the vehicles are a fleet of transport vehicles. In some cases, the vehicles are aerial drones. In some cases, the vehicles are marine drones. In some cases, the vehicles are shipping vessels. In some cases, the vehicles are trains. In some cases, the vehicles are trucks, such as transport trucks. In some cases, the vehicles are space vehicles, such as satellites.Atty. Dkt. No. 142937-5002

[0068] FIG. 3 shows another example of a computing system 300 showing external data sources 302 and client data sources 304 that provide data that is ingested by another embodiment of an Al computing environment 310. In some cases, the Al computing environment 310 executes the operations described with respect to the Al computing environment 110, including computations of a client-dedicated Al computing environment (e.g., 118, 124).

[0069] In some cases, the Al computing environment includes one or more client Al agents 312, which are part of one or more data pipelines. The one or more data pipelines include: pattern recognition and anomaly detection, Al models, vendor insights, a labeling module, training data, a predictive analytics module, a risk assessment module, a compliance module, a vendor vector store, an ontology / taxonomy of keywords and concepts, or a module for adjustable self-tuning of weights and bias, or one or more various combinations thereof.

[0070] In some cases, a data pipeline includes both a forecast module configured to generate forecasts and a promotion module configured to suggest autonomous product, price, place, and promotion (4Ps) recommendations. In some cases, the promotion module autonomously analyses firmographic, demographic, and exogenous data, and corresponding forecasted data from the forecast module, to generate the 4Ps recommendations. In some cases, the forecast module computes forecasts for stock and the promotion module uses the forecasts to generate a new old stock (NOS) autonomous recommendation to sell inventory that is new but old and apply the 4Ps to sell and get the inventory off the books. In some cases, the forecast module includes a look-a-like I cosine similarity sub-module to model and forecast new products using the 4Ps recommendations to help model and forecast new product introductions based on prior customer, vendor, firmographic and demographic similar historical structured and unstructured data and machine learnings.

[0071] In some cases, the client Al agent 312 generates outputs 320, including: a score, a Strength Weakness Opportunities Threats (SWOT) analysis, an explanation for the score, an explanation for the SWOT analysis, or a recommendation, or one or more various combinations thereof. In some cases, one or some of these outputs 320 are ingested by the client Al agent 312 for additional processing.Atty. Dkt. No. 142937-5002

[0072] FIG. 4 shows another computing system 400 that is used for rapidly training and deploying Al agents. In some cases, the computing system 400 executes the operations described with respect to the Al computing environment 110, including computations of a client-dedicated Al computing environment (e.g., 118, 124). The computing system 400, in some cases, is a stack of module that includes underlying infrastructure 402, data storage 404, core 406 (e.g., data pipelines), multiple client Al agents 408 (e.g., across retail industry, logistics industry, cybersecurity industry, etc.), and a user layer 410 (e.g., including users interacting directly with client Al agents, or an Al agent interfacing with a client Al agent, or an application programming interface (API) interfacing with a client Al agent, or a combination thereof).

[0073] FIG. 5 shows executable operations by training.

[0074] In some cases, blocks 504 to 510 are executed in an Al computing environment that is external to a client Al computing environment. In some cases, blocks 512 to 522 are executed in a client Al computing environment.

[0075] Block 504: Train a fundamental Al model using a plurality of analytics across a plurality of industries.

[0076] Block 506: For a given industry, the computing system 110 ingests industry-specific data from one or more external data sources into an industry database.

[0077] Block 508: For the given industry, the computing system 110 computes synthetic industry-specific data and augment the industry-specific data in the industry specific database.

[0078] Block 510: For the given industry, the computing system 110 re-trains the fundamental Al model using the industry-specific data to generate an industryspecific Al model.

[0079] Block 512: Deploy the trained industry- specific Al model to a client computing system (e.g., 118 for Client 1 ).

[0080] Block 514: The client computing system ingests client-specific data from one or more client data sources into a client database.

[0081] Block 516: The client computing system re-trains the industry-specific Al model using the client data to generate a trained client Al model.Atty. Dkt. No. 142937-5002

[0082] Block 518: The client computing system re-deploys the trained client Al model at the client.

[0083] Block 520: The client computing system ingests new client-specific data from the one or more client data sources into the client database.

[0084] Block 522: The client computing system re-trains the client Al model using the new client data to generate a re-trained version of the client Al model, and re-deploy

[0085] The process of blocks 520 and 522 are repeated, generating new versions of the client Al model over time.

[0086] Referring to FIG. 6, an example process is provided for executing a train client Al model.

[0087] Block 602: During training, ingest control data (e.g., PDFs, text documents, presentation slides, spreadsheets, emails, videos, audio data, etc.) comprising natural language into the client Al model to determine a plurality of natural language logic conditions. In some cases, these natural language logic conditions are called prompts.

[0088] In some cases, humans (e.g., users of the computing system) provide natural language data (e.g., via voice or text data) to train and refine the client Al model.

[0089] Block 604: After training, at a first data node, ingest unstructured data (e.g., natural language text data or voice data (or both), PDFs, text documents, presentation slides, emails, screenshots, images, video, audio, graphs, etc.) and / or structured data (e.g., machine data, tables, structured data files, spreadsheets, transaction data, etc.) into the client Al model. In some cases, the natural language text data or voice data (or both) is provided via a chatbot interface.

[0090] Block 606: Use the client Al model to execute an analysis to determine a first data node score relevant to one or more of the natural language logic conditions.

[0091] Block 608: After training, at a second data node, ingest unstructured data (e.g., natural language text data or voice data (or both), PDFs, text documents, presentation slides, emails, screenshots, video, audio, images, etc.) and / or structured data (e.g., machine data, tables, structured data files, spreadsheets, etc.) into theAtty. Dkt. No. 142937-5002 client Al model. In some cases, the natural language text data or voice data (or both) is provided via a chatbot interface.

[0092] Block 610: Use the client Al model to execute an analysis to determine a second data node score (e.g., for compliance and / or non-compliance, SWOT analysis, recommendations, etc.) relevant to one or more of the natural language logic conditions.

[0093] Block 612: In some cases, the client Al agent automatically generates a report (or a message or platform output) summarising the scores and reasons based on the natural language logic conditions and contextual information obtained from the ingested control documents at block 602.

[0094] In some cases, the client Al agent automatically transmits and receives messages between other internal Al agents and / or other external Al agents (associated with and / or operated by customers, manufacturers, vendors) in the industry ecosystem. These messages include, in some cases, the scores and reasons based on the natural language logic conditions and contextual information obtained from the ingested control documents

[0095] Block 614: The client Al agent determines if an urgent condition outputted is by the client Al model based on one or more of the scores. If so, then the process continues to block 616.

[0096] Block 616: Generate and send an alert to one or more client devices comprising natural language that describes a reason for compliance and / or non- compliance based on the natural language logic conditions and contextual information obtained from the ingested control documents at block 602

[0097] In some cases, an automated alert is transmitted to other internal Al agents and or external Al agents to take specific actions based on policies, procedures, business rules, metrics, KPIs, events, thresholds, etc.

[0098] FIG. 7 shows an example GUI 700 that facilities a user to upload control data files into the Al computing environment 110 for training a client Al model or for training an industry application model, or both. The user, for example, can drag and drop files for easy uploading.Atty. Dkt. No. 142937-5002

[0099] FIG. 8 shows an example GUI 800 that is displayable at the data source nodes, which facilities a user to upload data files for processing by a client Al agent. For example, users do not need to have expertise in computing (e.g., data pipelines, API, advanced data platforms). In some cases, users log into a portal that accesses the client Al agent, and can drag and drop data from the given data for processing by the client Al agent. For example, a user can take a photo of an invoice or a packing slip and upload the photo via the GUI 800. In another example, a PDF report generated by a node’s system can be uploaded via the GUI 800.

[0100] In some cases, a video file or sound file, or both, are uploaded via the GUI 800 for processing and training. In some cases, a data file arriving from loT devices, robots, vehicles, and machines can be uploaded via the GUI 800 for processing and training.

[0101] FIG. 9 shows that data may flow through the data node using one or more approaches, including a drag and drop GUI (e.g., such as GUI 800) and / or a data collection server. The data collection server obtains data from one or more data platforms or devices. Examples of data platforms and devices include one or a combination of: cameras, microphones, sensors, kiosks, and industrial equipment. This data is transmitted to the data collection server. In some cases, a data pipeline between the data collection server and the data node automatically facilitates the transmission of the data to the data node, and in turn to the client Al agent. In some cases, a user or an Al bot facilitates the transmission of the data from the data collection server to the drag and drop GUI, which is then sent to the client Al agent.

[0102] Referring to FIG. 10, there is illustrated a simplified block diagram of a computer in accordance with at least some embodiments. Computer 1000 is an example implementation of a computer for the Al computing environment 110, the client Al computing environment 118, the data nodes, the Al computing environment 310, the computing system 400, and for the other computing devices and servers. Computer 1000 has at least one processor 1010 operatively coupled to at least one memory 1020, at least one communications interface 1030 (also herein called a network interface), and at least one input / output device 1040.

[0103] The at least one memory 1020 includes a volatile memory that stores instructions executed or executable by processor 1010, and input and output dataAtty. Dkt. No. 142937-5002 used or generated during execution of the instructions. Memory 1020 may also include non-volatile memory used to store input and / or output data - e.g., within a database - along with program code containing executable instructions.

[0104] Processor 1010 may transmit or receive data via communications interface 1030, and may also transmit or receive data via any additional input / output device 440 as appropriate.

[0105] In some cases, the processor 1010 includes a system of central processing units (CPUs) 1012. In some other cases, the processor includes a system of one or more CPUs and one or more Graphical Processing Units (GPUs) 1014 that are coupled together. Although not shown, in combination or in alternative to GPUs, the CPUs are also coupled with tensor processing unit (TPUs), and / or other Al- dedicated processors. In some cases, the GPUs, TPUs and / or other Al-dedicated processors execute machine learning computations or neural network computations, or both. In some cases, the GPUs, TPUs and / or other Al-dedicated processors execute the operations of the Al agents described herein.

[0106] Various systems or processes have been described to provide examples of embodiments of the claimed subject matter. No such example embodiment described limits any claim and any claim may cover processes or systems that differ from those described. The claims are not limited to systems or processes having all the features of any one system or process described above or to features common to multiple or all the systems or processes described above. It is possible that a system or process described above is not an embodiment of any exclusive right granted by issuance of this patent application. Any subject matter described above and for which an exclusive right is not granted by issuance of this patent application may be the subject matter of another protective instrument, for example, a continuing patent application, and the applicants, inventors or owners do not intend to abandon, disclaim or dedicate to the public any such subject matter by its disclosure in this document.

[0107] For simplicity and clarity of illustration, reference numerals may be repeated among the figures to indicate corresponding or analogous elements. In addition, numerous specific details are set forth to provide a thorough understanding of the subject matter described herein. However, it will be understood by those of ordinary skill in the art that the subject matter described herein may be practicedAtty. Dkt. No. 142937-5002 without these specific details. In other instances, well-known methods, procedures, and components have not been described in detail so as not to obscure the subject matter described herein.

[0108] The terms “coupled” or “coupling” as used herein can have several different meanings depending in the context in which these terms are used. For example, the terms coupled or coupling can have a mechanical, electrical or communicative connotation. For example, as used herein, the terms coupled or coupling can indicate that two elements or devices are directly connected to one another or connected to one another through one or more intermediate elements or devices via an electrical element, electrical signal, or a mechanical element depending on the particular context. Furthermore, the term “operatively coupled” may be used to indicate that an element or device can electrically, optically, or wirelessly send data to another element or device as well as receive data from another element or device.

[0109] As used herein, the wording “and / or” is intended to represent an inclusive-or. That is, “X and / or Y” is intended to mean X or Y or both, for example. As a further example, “X, Y, and / or Z” is intended to mean X or Y or Z or any combination thereof.

[0110] Terms of degree such as "substantially", "about", and "approximately" as used herein mean a reasonable amount of deviation of the modified term such that the result is not significantly changed. These terms of degree may also be construed as including a deviation of the modified term if this deviation would not negate the meaning of the term it modifies.

[0111] Any recitation of numerical ranges by endpoints herein includes all numbers and fractions subsumed within that range (e.g., 1 to 5 includes 1 , 1.5, 2, 2.75, 3, 3.90, 4, and 5). It is also to be understood that all numbers and fractions thereof are presumed to be modified by the term "about" which means a variation of up to a certain amount of the number to which reference is being made if the result is not significantly changed.

[0112] Some elements herein may be identified by a part number, which is composed of a base number followed by an alphabetical or subscript-numerical suffix (e.g., 312a, or 312b). All elements with a common base number may be referred to collectively or generically using the base number without a suffix (e.g., 312).Atty. Dkt. No. 142937-5002

[0113] The systems and methods described herein may be implemented as a combination of hardware or software. In some cases, the systems and methods described herein may be implemented, at least in part, by using one or more computer programs, executing on one or more programmable devices including at least one processing element, and a data storage element (including volatile and non-volatile memory and / or storage elements). These systems may also have at least one input device (e.g. a pushbutton keyboard, mouse, a touchscreen, and the like), and at least one output device (e.g. a display screen, a printer, a wireless radio, and the like) depending on the nature of the device. Further, in some examples, one or more of the systems and methods described herein may be implemented in or as part of a distributed or cloud-based computing system having multiple computing components distributed across a computing network. For example, the distributed or cloud-based computing system may correspond to a private distributed or cloud-based computing cluster that is associated with an organization. Additionally, or alternatively, the distributed or cloud-based computing system be a publicly accessible, distributed or cloud-based computing cluster, such as a computing cluster maintained by Microsoft Azure™, Amazon Web Services™, Google Cloud™, or another third-party provider. In some instances, the distributed computing components of the distributed or cloudbased computing system may be configured to implement one or more parallelized, fault-tolerant distributed computing and analytical processes, such as processes provisioned by an Apache Spark™ distributed, cluster-computing framework or a Databricks™ analytical platform. Further, and in addition to the CPUs described herein, the distributed computing components may also include one or more graphics processing units (GPUs) capable of processing thousands of operations (e.g., vector operations) in a single clock cycle, and additionally, or alternatively, one or more tensor processing units (TPUs) capable of processing hundreds of thousands of operations (e.g., matrix operations) in a single clock cycle.

[0114] Some elements that are used to implement at least part of the systems, methods, and devices described herein may be implemented via software that is written in a high-level procedural language such as object-oriented programming language. Accordingly, the program code may be written in any suitable programming language such as Python or Java, for example. Alternatively, or in addition thereto, some of these elements implemented via software may be written in assemblyAtty. Dkt. No. 142937-5002 language, machine language or firmware as needed. In either case, the language may be a compiled or interpreted language.

[0115] At least some of these software programs may be stored on a storage media (e.g., a computer readable medium such as, but not limited to, read-only memory, magnetic disk, optical disc) or a device that is readable by a general or special purpose programmable device. The software program code, when read by the programmable device, configures the programmable device to operate in a new, specific, and predefined manner to perform at least one of the methods described herein.

[0116] Furthermore, at least some of the programs associated with the systems and methods described herein may be capable of being distributed in a computer program product including a computer readable medium that bears computer usable instructions for one or more processors. The medium may be provided in various forms, including non-transitory forms such as, but not limited to, one or more diskettes, compact disks, tapes, chips, and magnetic and electronic storage. Alternatively, the medium may be transitory in nature such as, but not limited to, wire-line transmissions, satellite transmissions, internet transmissions (e.g., downloads), media, digital and analog signals, and the like. The computer usable instructions may also be in various formats, including compiled and non-compiled code.

[0117] While the above description provides examples of one or more processes or systems, it will be appreciated that other processes or systems may be within the scope of the accompanying claims.

[0118] To the extent any amendments, characterizations, or other assertions previously made (in this or in any related patent applications or patents, including any parent, sibling, or child) with respect to any art, prior or otherwise, could be construed as a disclaimer of any subject matter supported by the present disclosure of this application, Applicant hereby rescinds and retracts such disclaimer. Applicant also respectfully submits that any prior art previously considered in any related patent applications or patents, including any parent, sibling, or child, may need to be revisited.

Claims

Atty. Dkt. No. 142937-5002What is claimed is:1 . A computing system for training artificial intelligence (Al) models, the computing system comprising a processor, a communication interface, and memory, the processor coupled to the communication interface and the memory; the memory comprising a fundamental Al model that is trained; and the processor configured to at least: ingest, from an external data source, industry-specific data; re-train the fundamental Al model using the industry-specific data to generate an industry-specific Al model; deploy the industry-specific Al model; ingest, from a client data source, client data; re-train the industry-specific Al model using the client data to generate a client Al model; deploy the client Al model; ingest, from the client data source, new client data; re-train the client Al model using the new client data to generate a new client Al model, and deploy the new client Al model in place of the client Al model.

2. The computing system of claim 1 , further comprising a client Al computing environment and an Al computing environment separate from the client Al computing environment; wherein the industry-specific Al model is deployed within the client Al computing environment; wherein the client Al model is generated and deployed within the client Al computing environment; wherein the new client Al model is generated and deployed within the client Al computing environment; and wherein the industry-specific Al model is generated in the Al computing environment.

3. The computing system of claim 2, wherein the client Al computing environment comprises one or more security permissions to correspond with the client data source; and wherein and the one or more security permissions, the client data source, the client Al model, and the new client Al model are inaccessible by the client Al computing environment.Atty. Dkt. No. 142937-50024. The computing system of claim 2, further comprising a second client Al computing environment that is separate from the client Al computing environment and the Al computing environment; the processor further configured to: deploy the industry-specific Al model in the second client Al computing environment; ingest, from a second client data source accessible by the second client Al computing environment, second client data; re-train the industry-specific Al model using the second client data to generate a second client Al model; deploy the second client Al model in the second client Al computing environment; ingest, form the second client data source, new second client data; re-train the second client Al model using the new second client data to generate a new second client Al model, and deploy the new second client Al model in place of the second client Al model in the second client Al computing environment.

5. The computing system of claim 2, wherein the client data source is part of a client system, and the client Al computing environment comprises a digital container that establishes secure communication with the client system.

6. The computing system of claim 1 , wherein the processor is further configured to: after deploying the new client Al model, ingest, from the external data source, new industry-specific data; re-train the fundamental Al model or the industry-specific Al model, or both, using the using the new industry-specific data to generate a new industry-specific Al model; deploy the new industry-specific Al model; re-train the new industry-specific Al model using the new client data to generate a second new client Al model; and deploy the second new client Al model in place of the new client Al model.

7. The computing system of claim 1 , wherein ingesting the client data comprises: the computing system communicating with multiple data nodes of the client data source;Atty. Dkt. No. 142937-5002 wherein the multiple data nodes comprises: a first data node that generates a first output; a second data node that is configured to process at least the first output to generate a second output; and a third data node that is configured to process at least the second output to generate a third output; and, wherein the first output, the second output and the third output form at least part of the client data.

8. The computing system of claim 1 , wherein ingesting the client data comprises: the computing system communicating with multiple data nodes of the client data source; and wherein one or more of the multiple data nodes comprises a node Al agent that is configured to generate node data that forms at least part of the client data.

9. The computing system of claim 8, wherein the multiple data nodes comprise multiple robots.

10. The computing system of claim 8, wherein the multiple data nodes comprise multiple vehicles.11 . The computing system of claim 8, wherein the multiple data nodes comprise multiple machines.

12. The computing system of claim 1 , wherein, the new client Al model communicates with a client computing device, and, when the new client Al model is deployed, the new client Al model is configured to generate a message that is transmitted to the client computing device; and, wherein the message comprises an analysis of at least the client data and the new client data.

13. The computing system of claim 12, wherein the message is automatically generated in response to the computing system detecting a query from the client computing device.Atty. Dkt. No. 142937-500214. The computing system of claim 12, wherein the message is automatically generated in response to the computing system detecting new industry specific data that differs from the industry-specific data.

15. The computing system of claim 12, wherein the message is automatically generated in response to the computing system detecting additional new client data that differs from the new client data and the client data.

16. The computing system of claim 1 , wherein the processor is further configured to at least: ingest control data comprising natural language; process the control data using the client Al model or the new client Al model to generate a plurality of natural language logic conditions; when the client Al model or the new client Al model is deployed, receive the client data from a data node that is part of the client data source; determine, using the client Al model, a node data score relevant to one or more of the natural language logic conditions; and determine if the node data score triggers an urgent condition, and, responsive to triggering the urgent condition, generating and sending a message comprising information about the data node.

17. The computing system of claim 16, wherein the message comprises a natural language description regarding the data node, contextual information obtained from the control data, and compliance or non-compliance based on the one or more of the natural language logic conditions.

18. The computing system of claim 1 , wherein, when ingesting the client data from the client data source, the processor is configured to provide a graphical user interface comprising a drag-and-drop data collection object; wherein the drag-and- drop data collection object is configured to receive one or more data files via a drag- and-drop interaction, the one or more data files comprising the client data.

19. A method for training and deploying artificial intelligence (Al) models, the method executed in a computing environment comprising a processor, a communicationAtty. Dkt. No. 142937-5002 interface, and memory, the processor coupled to the communication interface and the memory, and the method comprising: storing a fundamental Al model that is trained in the memory ingesting, from an external data source, industry-specific data; re-training the fundamental Al model using the industry-specific data to generate an industry-specific Al model; deploying the industry-specific Al model; ingesting, from a client data source, client data; re-training the industry-specific Al model using the client data to generate a client Al model; deploying the client Al model; ingesting, from the client data source, new client data; and re-training the client Al model using the new client data to generate a new client Al model, and deploying the new client Al model in place of the client Al model.

20. A non-transitory computer readable medium storing computer executable instructions which, when executed by at least one computer processor, cause the at least one computer processor to carry out a method for training and deploying artificial intelligence (Al) models, the method comprising: storing a fundamental Al model that is trained in a memory system in communication with the at least one computer processor; ingesting, from an external data source, industry-specific data; re-training the fundamental Al model using the industry-specific data to generate an industry-specific Al model; deploying the industry-specific Al model; ingesting, from a client data source, client data; re-training the industry-specific Al model using the client data to generate a client Al model; deploying the client Al model; ingesting, from the client data source, new client data; and re-training the client Al model using the new client data to generate a new client Al model, and deploying the new client Al model in place of the client Al model.

Images