Systems and methods for providing adaptive AI-driven conversational agents

Abstract

A system and method for providing adaptive and interactive AI-driven profiles ingests branded content data, organizes the data into embeddings and indexes, stores the plurality of embeddings and indexes in a knowledge base, generates a user profile based on the organized embeddings and indexes and a user history of the user associated with the user profile, updates the user profile based on one or more models trained based on records indicative of interactions between the user and the user profile and one or more actions of the human user, personalizes responses of a conversational agent interacting with the first user based on the first user profile, provides customized content recommendations for the user, and further provides data-driven recommendations regarding improvement of the responses of the respective conversational agent, improvement of one or more services provided, and system performance.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to Provisional Application No. 63 / 448,117, filed February 24, 2023, the entire subject matter of which is incorporated herein by reference. [Background technology]

[0002] The present disclosure relates generally to artificial intelligence systems, and more particularly to improving user engagement in artificial intelligence-driven communications.

[0003] Current artificial intelligence (AI) systems are limited in their ability to understand and respond to each user's unique needs and preferences. While AI can process vast amounts of data and information, it lacks the ability to personalize the characteristics of the responses it generates. Without personalization, it is impossible to build relationships and memories with users. As a result, existing AI systems often fail to meet user needs by taking a generic, one-size-fits-all approach to content creation and delivery, resulting in limited and ineffective guidance and recommendations that fail to create meaningful connections with users. Summary of the Invention

[0004] Aspects of the present disclosure relate to methods, apparatus, and / or systems for providing adaptive AI-driven conversational agents. [Means for solving the problem]

[0005] In some aspects, the technology described herein relates to a method for generating an adaptive and interactive AI-driven profile, the method comprising: ingesting, by a processor, a first branded content dataset; organizing, by the processor, the ingested first branded content dataset into a plurality of embeddings and indexes; and storing the plurality of embeddings and indexes in a knowledge base, where the embeddings include embeddings of vectors in an embedding space, the location of the embeddings providing a semantic meaning of the content represented by the vectors; and the index includes a data structure providing a mapping between the branded content data and its location in the knowledge base and a link to metadata associated with the content data; the method further comprising: generating a first user profile. The method includes: generating, by a processor, a plurality of organized embeddings and indexes based at least in part on a user history of a first user associated with a first user profile; updating, by the processor, the first user profile based at least in part on one or more models trained based on recordings indicative of one or more interactions between the first user and the first user profile and one or more actions of the human user; and personalizing, by the processor, one or more responses of a conversational agent interacting with the first user based at least in part on the first user profile.

[0006] In some embodiments, in a method related to the technology described herein, ingesting the first branded content dataset includes processing the first branded content dataset with one or more machine learning algorithms and identifying one or more insights about the first branded content dataset.

[0007] In some embodiments, in methods involving the technology described herein, the one or more insights include at least one of tone, language, audience engagement, intent, mood, receptivity, skill, expertise, or understanding.

[0008] In some embodiments, the methods of the technology described herein involve one or more models trained based on recordings indicative of one or more processing of a human user, including one or more models trained based on recordings indicative of biological-cognitive and neuroscientific processing of the human user, the models providing information about at least one of the user's thought processes, behavioral patterns, motivations, or biases.

[0009] In some embodiments, in methods involving the technology described herein, the first user profile is updated in real time.

[0010] In some aspects, a method involving the technology described herein further comprises generating one or more customized content recommendations for the first user based at least in part on the first user profile, and providing, by a conversation agent, the one or more customized content recommendations to the first user.

[0011] In some embodiments, a method involving the technology described herein further comprises: analyzing, by a processor, input from a plurality of users in response to interactions with a respective conversational agent; extracting, by the processor, one or more insights related to the interactions with the plurality of users; and providing, by the processor, one or more data-driven recommendations related to at least one of improving the responses of the respective conversational agent, improving one or more services provided, or system performance.

[0012] In some aspects, the technology described herein relates to a system for generating adaptive and interactive AI-driven profiles, the system comprising: a computer having a processor and a memory; and one or more sets of code stored in the memory and executed by the processor, the one or more sets of code configured to: ingest a first branded content data set; organize the ingested first branded content data set into a plurality of embeddings and indexes; and store the plurality of embeddings and indexes in a knowledge base, where the embeddings include embeddings of vectors in an embedding space, the locations of the embeddings confer semantic meaning of the content represented by the vectors; and the indexes include a mapping between the branded content data and the locations of the branded content data in the knowledge base and a mapping between the content data and the locations of the branded content data in the knowledge base and a mapping between the content data and the locations of the branded content data in the knowledge base. and a link to the associated metadata, wherein the processor is further configured to: generate a first user profile based at least in part on the organized plurality of embeddings and indexes and a user history of the first user associated with the first user profile; update the first user profile based at least in part on one or more models trained based on recordings indicative of one or more interactions between the first user and the first user profile and one or more actions of the human user; and personalize one or more responses of a conversational agent interacting with the first user based at least in part on the first user profile.

[0013] In some embodiments, in a system to which the technology described herein relates, ingesting the first branded content dataset includes processing the first branded content dataset with one or more machine learning algorithms and identifying one or more insights about the first branded content dataset.

[0014] In some embodiments, in systems to which the technology described herein relates, the one or more insights include at least one of tone, language, audience engagement, intent, mood, receptivity, skill, expertise, or understanding.

[0015] In some embodiments, in systems to which the technology described herein relates, the one or more models trained based on recordings indicative of one or more processes of a human user include one or more models trained based on recordings indicative of biological-cognitive and neuroscientific processes of the human user that provide information about at least one of the user's thought processes, behavioral patterns, motivations, or biases.

[0016] In some embodiments, in systems to which the technology described herein pertains, the first user profile is updated in real time.

[0017] In some aspects, the technology described herein relates to a system further configured to generate one or more customized content recommendations for the first user based at least in part on the first user profile, and provide, by a conversation agent, the one or more customized content recommendations to the first user.

[0018] In some embodiments, a system to which the technology described herein relates is further configured to analyze input from a plurality of users in response to interactions with a respective conversational agent, extract one or more insights related to the interactions with the plurality of users, and provide one or more data-driven recommendations regarding one or more of improving the responses of the respective conversational agent, improving one or more services provided, or system performance.

[0019] In some aspects, the technology described herein relates to a non-transitory computer-readable medium storing computer program instructions that, when executed by one or more processors, cause the one or more processors to perform a process including ingesting a first branded content data set, organizing the ingested first branded content data set into a plurality of embeddings and indexes, and storing the plurality of embeddings and indexes in a knowledge base, where the embeddings include embeddings of vectors in an embedding space, the locations of the embeddings confer semantic meaning of the content represented by the vectors, and the index includes a data structure that provides a mapping between the branded content data and the location of the branded content data in the knowledge base and a link to metadata associated with the content data; The one or more processors are further caused to perform processes including generating a first user profile based at least in part on the organized plurality of embeddings and indexes and a user history of the first user associated with the first user profile; updating the first user profile based at least in part on one or more models trained based on recordings indicative of one or more interactions between the first user and the first user profile and one or more actions of the human user; and personalizing one or more responses of a conversational agent interacting with the first user based at least in part on the first user profile.

[0020] In some embodiments, in a non-transitory computer-readable medium related to the technology described herein, ingesting the first branded content dataset includes processing the first branded content dataset with one or more machine learning algorithms and identifying one or more insights about the first branded content dataset.

[0021] In some embodiments, in a non-transitory computer-readable medium to which the technology described herein relates, the one or more insights include at least one of tone, wording, audience engagement, intent, mood, receptivity, skill, expertise, or understanding.

[0022] In some embodiments, the technology described herein relates to a non-transitory computer-readable medium, wherein the one or more models trained based on recordings indicative of one or more processing of a human user include one or more models trained based on recordings indicative of biological-cognitive and neuroscientific processing of the human user, the models providing information about at least one of the user's thought processes, behavioral patterns, motivations, or biases.

[0023] In some aspects, the technology described herein relates to a non-transitory computer-readable medium, further comprising generating one or more customized content recommendations for the first user based at least in part on the first user profile, and providing the one or more customized content recommendations to the first user by a conversation agent.

[0024] In some aspects, the technology described herein relates to a non-transitory computer-readable medium, further comprising: analyzing, by a processor, input from a plurality of users in response to interactions with a respective conversational agent; extracting, by the processor, one or more insights related to the interactions with the plurality of users; and providing, by the processor, one or more data-driven recommendations related to one or more of improving the responses of the respective conversational agent, improving one or more services provided, or system performance.

[0025] Various other aspects, features, and advantages will become apparent by reference to the detailed description and accompanying drawings. Also, it is to be understood that both the foregoing general description and the following detailed description are exemplary only and are not intended to limit the scope of the present disclosure. [Brief explanation of the drawings]

[0026] [Figure 1] 1 illustrates an exemplary system for providing an adaptive AI-driven conversational agent in accordance with at least one embodiment.

[0027] [Figure 2] FIG. 2 illustrates an example method for providing an adaptive AI-driven conversational agent according to at least one embodiment.

[0028] [Figure 3] 1 is an example of a user interface implementing an admin interface in accordance with at least one embodiment.

[0029] [Figure 4] 1 is an example of a user interface implementing a conversational agent in accordance with at least one embodiment.

[0030] [Figure 5] FIG. 1 is a physical architecture block diagram illustrating an example of a computing device (or data processing system) that may implement aspects of the described technology.

[0031] While the technology of the present invention is susceptible to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the drawings and are herein described in detail. The drawings may not be to scale. It should be understood, however, that the drawings and the detailed description thereunder are not intended to limit the technology to the particular forms disclosed, but on the contrary, are intended to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the technology as defined by the appended claims. DETAILED DESCRIPTION OF THE INVENTION

[0032] In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of various embodiments. However, those skilled in the art will understand that embodiments of the present invention may be practiced without these specific details or with equivalent configurations. In other instances, well-known structures and devices are shown in block diagram form in order to avoid unnecessarily obscuring embodiments of the present invention.

[0033] To mitigate the problems described herein, the inventors have had to devise solutions and, in some cases, just as importantly, recognize problems that others in the field of artificial intelligence have overlooked (or have not yet foreseen). Indeed, the inventors wish to emphasize the difficulty of recognizing problems in their early stages. These problems will become much more apparent in the future if industry trends continue as the inventors expect. Furthermore, because there are multiple problems addressed, it should be understood that some embodiments are specialized to one problem or another, and not all embodiments address all of the problems of conventional systems described herein or provide all of the advantages described herein. That is, improvements that solve various permutations of these problems are described below.

[0034] The disclosed techniques for personalizing AI response features offer significant improvements over conventional artificial intelligence (AI) systems. Embodiments of the disclosed AI systems employing the techniques disclosed herein have enhanced capabilities for generating AI responses that engage with users on a personal level. As discussed above, conventional AI systems are limited in their ability to understand and respond to the unique needs and preferences of individual users. For example, traditional chatbots are rule-based, following predetermined conversational flows and utilizing rigid "if X (condition) then Y (action)" formulas. Such rigid systems are limited to options coded during development, placing an extraordinary burden on developers even for relatively simple process flows. In some embodiments, employing AI technologies enables the use of more flexible architectures. For example, AI chatbots can employ conversational AI technologies that utilize advanced technologies such as natural language processing, natural language understanding, machine learning, deep learning, and predictive analytics to provide a more dynamic and less constrained user experience. However, even with these advances, conversational AI systems can still be limited by a lack of personalization (although this does not suggest that these technologies should be abandoned).

[0035] The limitations of AI systems with regard to personalization stem from several factors. First, most AI systems are based on existing datasets and algorithms designed to recognize patterns of general user behavior. This means they are unable to adapt to the unique circumstances and individual needs of different users. As a result, these systems can only provide generic, one-size-fits-all recommendations and guidance, unable to tailor them to the specific needs and preferences of users. Another factor is a lack of understanding of the motivations behind user behavior. AI systems cannot understand the underlying reasons why users behave in a certain way, which is crucial for providing a truly personalized experience. This limits their ability to understand each user's context and individual needs and provide accurate and meaningful recommendations and guidance. Finally, current AI systems lack the ability to remember users and build relationships with them. While AI systems can process vast amounts of data and information, they lack the ability to build unique relationships and memories for each user. This means they cannot adapt and evolve over time to a given user, let alone multiple different users, which is key to providing a truly personalized experience.

[0036] The disclosed technology ameliorates the above-mentioned deficiencies and limitations of conventional chatbots and conversational AI systems by providing a conversational agent that enhances current AI-generated response features, such as AI-generated responses that take into account user differences to provide a personalized user experience. Embodiments of the disclosed technology include, but are not limited to, improvements to conversational AI systems with components for brand learning, personalized rapport, adaptive curation, and intelligence and insights, as described herein. One or more of these various components can work together to create a highly customized and personalized user experience and provide accurate and meaningful guidance and recommendations to users. For example, one or more of these components can be incorporated as or into one or more trained AI models employed to generate responses in a conversational AI system in accordance with the disclosed technology, overcoming the limitations of current AI systems and bridging the gap between generic and personalized user experiences to provide new levels of engagement and relevance to users.

[0037] Embodiments of the disclosed technology improve the responsiveness of conversational AI systems for a wide range of use cases, taking personalized AI engagement to a new level. For example, consider an AI tutor that provides individualized instruction on any topic, adapting to each student's learning style and pace. Another application is entertainment, where celebrities or influencers can use embodiments described herein to deepen fan engagement through personalized conversations. In commerce, the disclosed technology may be used to revolutionize the personal shopping experience with an AI shopping assistant that learns a user's preferences over time and provides recommendations aligned with those preferences. In the realm of causes and campaigns, the disclosed technology may offer new ways to engage users by understanding their motivations and providing personalized messaging and calls to action. These are just a few of the many potential applications of the systems and methods described herein and their ability to take personalized AI engagement to a new level.

[0038] FIG. 1 illustrates an exemplary system 100 for providing an adaptive AI-powered conversational agent according to at least one embodiment. As described above, the personalization-enhanced conversational AI system 100 may include one or more components, such as a brand learning module 110, a personalized rapport module 120, an adaptive curation module 130, and an intelligence and insights module 140, as described in detail herein. In some examples, one or more of these components may be combined into a suite of advanced artificial intelligence capabilities to engage with end users 150 in new and exciting ways. In some examples, these components may exist as one or more trained models, including AI models. In some examples, the functionality of these components may be mixed, such as within a single model, or used in a pipeline of multiple different models. In various embodiments, these and other modules may be implemented to provide an adaptive AI-powered conversational agent that may be configured to interact with end users 150, as described in detail herein.

[0039] As used herein, a conversational agent is understood to refer to an interaction system that performs natural language processing (NLP) and automatically responds using human language. A conversational agent represents an implementation of computational linguistics and, in various embodiments, may be deployed as a chatbot, a virtual assistant, an AI assistant, or the like. Conversational agents may be implemented on various platforms, such as messaging apps, websites, or standalone applications, and may be employed to provide information, answer questions, perform tasks, or assist users in achieving specific goals. Conversational agents can enhance the user experience by facilitating seamless interaction between humans and machines and providing an accessible and efficient means of communication.

[0040] In some embodiments, the brand learning module 110 may correspond to a deep training component that ingests content (e.g., proprietary content, etc.) such as articles, videos, podcasts, and the like and continuously evolves to learn the essence and DNA of a brand. For example, the deep training component continuously evolves (potentially in real time) to understand the essence and DNA of a brand and serves as the foundation for other components. In some embodiments, the personalized rapport module 120 may correspond to a cognitive engagement component that leverages advances in cognition and neuroscience to learn from users, actively engage with them, and create lasting connections. In some embodiments, the adaptive curation module 130 may correspond to a component that dynamically creates and delivers individualized content tailored to each user's preferences and needs across various platforms and media based on personalized recommendations driven by advanced machine learning analysis of user interactions. In some embodiments, the intelligence and insights module 140 may correspond to a reporting, insights, and recommendations component that delivers deep insights and recommendations based on continuous machine learning analysis of conversations between an AI and a user. This suite of AI capabilities offers new and creative ways to interact with users, delivering a variety of valuable experiences that were previously unavailable.

[0041] In some embodiments, the conversational AI system 100 may include multiple components. For example, the conversational AI system 100 may include one or more components for brand learning, personalized rapport, adaptive curation, and intelligence and insights, among others. In some examples, the conversational AI system 100 may use one or more language models to which one or more other components interface, such as to provide input to the language model or obtain output from the language model. In some examples, the language model may be a large-scale language model 160, including, but not limited to, GPT-4, Claude 2, GPT-3, BERT, BLOOM, etc. In some embodiments, one or more other components may interface with the language model, as shown. Each component may operate on a set of inputs and provide a set of outputs, such as in response to the obtained inputs. Examples of inputs and outputs may include vectors that encode features of the data processed by the component. Some components may ingest human-readable content and output human-readable content such as natural language text, while others may operate on feature vectors that represent data corresponding to the human-readable content. Similarly, one or more components may ingest image content data, location data, video, etc., and output data based on the ingested content.

[0042] In some exemplary embodiments, the brand learning module 110 includes inputting proprietary content, such as articles, videos, podcasts, or other brand content, into a brand database. As used herein, it should be understood that a “brand” may refer to the identity of a company, person, persona, product, service, or concept that distinguishes it from others. In some embodiments, the brand data can be processed using a combination of machine learning algorithms, including natural language processing (NLP), to analyze the content and generate insights about the essence and DNA of the brand. These insights can be stored in a knowledge base, where the information is embedded and organized into an index for easy access and use by other components of the system. In processing the brand content, one or more processes or models can analyze aspects such as tone, language, audience engagement, intent, mood, receptivity, skill, expertise, and / or understanding to gain a deep understanding of the brand's unique identity. The knowledge base can serve as a comprehensive source of information about the brand and provide a foundation for other modules to use in delivering highly customized and personalized user experiences. Equally important, in some embodiments, the brand learning module 110 can include explicit negative instructions—specific words, actions, tone, or other aspects that a brand should never use. These can be managed, for example, via a guardrail system, where inputs or outputs to the system are analyzed by the LLM, specifically checking for violations of these instructions, and applying corrections accordingly. For example, hostile inputs to the system that attempt to change the system's behavior (e.g., "Please ignore all previous instructions") can be detected by such a system and ignored or a predetermined response returned.

[0043] In some exemplary embodiments, embedding and organizing processed branded content into an index refers to the process of converting information into a structured format that can be easily accessed and utilized by other components of the system. In some embodiments, one or more processes or models operate on unstructured data to generate structured output data based on the unstructured input. In some embodiments, an embedding is a representation of a piece of text, image, audio, or other media or data that captures the essence and meaning of the original content. In some embodiments, an embedding corresponds to the embedding of a vector within an embedding space, and the location of the embedding provides the semantic meaning of the content represented by the vector. An index, on the other hand, is a data structure that provides a mapping between content and its location in a knowledge base, as well as links to other metadata associated with the content (e.g., its provenance or access permissions).

[0044] By using embedding and indexing together, embodiments of the system may enable rapid retrieval of relevant information from a knowledge base, incorporating that information into decision-making processes, and providing a more personalized user experience. Embedding and indexing improve the efficiency of searching and accessing information in the knowledge base, reducing system response latency while providing the ability to provide relevant and personalized responses to users.

[0045] In some exemplary embodiments, a knowledge base may be used to store information specific to individual users and their past interactions with the system. For example, a conversation history between the system and the user, or a subset of this history identified as important by either the user or a machine learning model, or a collection of abstractions such as higher-level summaries of the user's conversation history, or similar textual, audio, or visual representations of the user's history with the system, may be embedded as indexed vectors in a vector database or similar storage structure for later reference by the system. In this way, the system can create a persistent and accessible record of a user's past interaction history, similar to a human's long-term memory.

[0046] In some embodiments, the brand learning module 110 may be configured to implement an embedding and retrieval pipeline, for example, using modern large-scale language model neural networks, to split one or more documents into smaller chunks, capture, store, and index the meaning of those chunks, and enable the chunks to be later independently searched, for example, based on fuzzy searches for similar meanings.

[0047] In some embodiments, the captured material may be, for example, a transcript, a PDF, a presentation, or any document or other media or data that consists of text or that includes text that can be extracted, for example, by transcription, computer vision, or other techniques. While the following description focuses on text as a use case, other embodiments may be directly applicable to other modalities, such as video or visual art (e.g., where the brand learning module 110 captures the visual content and style of an organization or individual rather than the written or spoken style and content). In such embodiments, embedding models designed for images, video, multimodal input, or other modes of input may be employed in place of the illustrated text embedding models, and while preprocessing steps may differ, the implementation may otherwise be substantially similar. These models may be configured to output points in vector space, with some or all of the accompanying functionality, as described herein.

[0048] In some embodiments, the text may be preprocessed, for example, by tokenization, case uniformity, spelling correction, stop-word removal, stemming, lemmatization, text normalization, or a subset of other techniques to standardize the text and increase information density. In some embodiments, the text is chunked, e.g., divided into overlapping sections consisting of N or fewer tokens or characters. In various embodiments, N can be varied to improve the overall performance of the brand learning embedding and retrieval pipeline, for example, but may be, for example, tens of tokens (e.g., generally words), hundreds of characters, a few complete sentences, or a single paragraph, or may be, for example, up to several hundred tokens.

[0049] In some embodiments, the processed text chunks can be passed through a neural network embedding model, such as one of the GTE (General Text Embeddings) family of open-source embedding models, the text-embedding-ada-002 model, its successors from OpenAI(), or one of many other commercial or open-source embedding models. These models take a stream of text as input and output a point in a high-dimensional vector space. The dimensionality of the space, as well as the chunk size, can be tuned to improve the performance of the embedding and search pipeline; for example, the space can have hundreds or thousands of dimensions. In some embodiments, the resulting vectors, the corresponding chunks of text, and other metadata, such as the original document, tags, upload date, and user data, can be stored in a dedicated vector database, such as Qdrant, Pinecone, Weaviate, or other commercial or open-source databases, or in more general databases that support vector searching, such as Redis or PostgreSQL.

[0050] In some embodiments, the chunks can then be searched in response to a natural language query. For example, the query can be processed in the same or similar manner as described herein (e.g., preprocessing the query and then running it through the same embedding model used to embed the original document), and the resulting vectors can be compared to vectors stored in a database using a similarity measure such as cosine distance. In some embodiments, search techniques can be implemented to find the most similar existing vectors in a space, including approximate nearest neighbor (ANN) and / or other vector search techniques, which are currently capable of identifying tens of nearest neighbor vectors out of millions in a few milliseconds, enabling real-time queries to support, for example, conversational agents. In some embodiments, the 5-500 closest matching neighbors can be re-ranked using different criteria, for example, using more complex algorithms that cannot be applied efficiently to millions of vectors.

[0051] In some embodiments, more advanced indexing techniques can be used, for example, by preprocessing each chunk in a different, more sophisticated way. For example, in some embodiments, each chunk can be run through a large-scale linguistic neural network model or other transformation tuned to summarize the meaning of the chunk or to generate a set of questions that the chunk may be important to answer. These summaries, questions, or other transformations can be embedded as described above, and each time the chunk may be indexed as an additional point in a vector space that can be compared to the query.

[0052] In some embodiments, an additional filtering step, e.g., on the metadata of each chunk, is performed before vector comparison, taking into account manual or automatic tags and other metadata along with the semantic and content of the text. For example, in some embodiments, this embedding-search pipeline can be applied to search expansion generation (RAG), whereby a targeted search against the embedded vector database is performed in response to a user query to generate a context for a conversational agent to generate a response, e.g., by inserting the retrieved text chunks into system prompts or messages that are used to generate the agent's response to the user.

[0053] Other embodiments may include, for example, running a "Codex Tools" query within or in addition to a system prompt to determine whether the agent should retrieve additional information from a vector database or use its internal memory and current context (e.g., system prompts and message history, etc.) to generate a response. In some embodiments, this query is editable by the agent creator. For example, the agent may be instructed to check whether the user is seeking information about insurance plans, and if so, to parse the question being asked and pass it to the vector database to search for and retrieve relevant information from a knowledge base. This retrieved information is passed to the system prompt or message history before generating a response to the user's query.

[0054] In some embodiments, executing the "Codex Tools" queries described herein enables one or more of the following:

[0055] Users can efficiently search for details in real time from this set of whitelisted documents (rather than relying, for example, on internal models of large language models or content surfaced by searching uncontrolled sources like the internet).

[0056] Users can get answers to their questions using the same methodology and documentation.

[0057] Users may be provided with direct citations and source data supporting the answers they receive.

[0058] Beyond simply defining source data, the conversational agent's responses may be controlled by the context obtained using the present RAG approach, e.g., updating its system prompts based on the type of query received.

[0059] The knowledge base that the conversational agent references and uses to answer questions and guide responses can be directly edited and managed by adjusting documents stored in the vector database, for example, to update insurance plan details or prices, or to update a person's tax code to the latest year.

[0060] In some example embodiments, the personalized rapport module 120 leverages information stored in the brand learning module's 110 knowledge base and the user's past interactions with the system to form lasting relationships with users. In some example embodiments, one or more processes or modules utilize machine learning algorithms, including those based on cognitive science and neuroscience, to continuously update user profiles based on each user's interactions, resulting in content and experiences that are personalized to each user's unique needs and preferences.

[0061] In some embodiments, the system may be provided with explicit, pre-set objectives or instructions to focus on specific topics or aspects during a conversation with the user. For example, the system may be instructed to specifically focus on gathering information related to the user's health and medical history. When conversing with the user, the system prioritizes remembering and building memories of health-related details and minimizes irrelevant details.

[0062] In another embodiment, the system allows for dynamic updating of memory objectives mid-conversation, allowing for real-time shifting of conversation focus and memory creation. For example, the system could start by collecting health-related memories, then shift to prioritize travel-related memories, while retaining previously collected health-related memories in a partitioned memory bank to prevent overwriting.

[0063] In some embodiments, explicit user-verified memory records may be generated for review, allowing users to directly review or correct their own memories. For example, after a health-focused conversation segment, the system may present extracted key medical memories for the user to verify accuracy, make edits or corrections, and ensure accurate representation before being permanently saved.

[0064] In some embodiments, different memory partitions are created to reflect different conversational purposes, dividing a user's memories into separate groups based on whether they relate to health, travel, education, or other topics. The purpose-specific memories can be efficiently referenced when needed.

[0065] In some embodiments, verified memories can be embedded as indexed vectors to enable fast context-based search and retrieval. For example, vector embeddings of knowledge graphs can be used to quickly identify health memories in the midst of relevant conversations.

[0066] In some embodiments, the validated user memory embedded as an indexed vector can be accessed by other conversational agents that have been granted explicit access permissions. This allows different agents to efficiently reference these memory vectors to enrich future conversations with that user. For example, an agent specialized in travel planning can access the user's health-related memory to better understand the user's medical needs and limitations in order to make travel recommendations. Access permissions are managed through a user profile database, allowing only authorized agents to access the user's memory-embedded vectors. This enables collaboration between agents, resulting in a seamless and personalized user experience across multiple conversation sessions.

[0067] In some exemplary embodiments, the personalized rapport module 120 uses a combination of cognitive and neuroscience-based principles and machine learning algorithms to continuously update a user profile to accurately reflect the user's evolving preferences and needs over time. Embodiments of this approach leverage one or more models trained on recordings indicative of a human user's biological cognitive and neuroscience processes to infer information about the user's thought processes and behavioral patterns, as well as their motivations and biases. For example, some cognitive processes may be characterized by a user's response time (e.g., reactive or deliberative), which may correspond to feedback signals obtained from user interactions during a conversation, such as dwell time before formulating a response / question, the time it takes the user to formulate a response / question, or user input corrections (e.g., total number of characters / words entered compared to number of characters / words submitted). Incorporating this understanding into the machine learning algorithms enables the system to build more meaningful and lasting relationships with users, rather than simply overwriting the profile with new information. The algorithmic embodiments employed can balance the importance of old and new information to provide a comprehensive understanding of the user and deliver a personalized experience that truly reflects the user's unique needs and preferences.

[0068] In some embodiments, a conversational agent that interacts with a user may be personalized for each user based on a number of different attributes that are stored and updated, for example, in a user profile object and provided to the agent at runtime. Some examples may include appropriate reading level and language, preferred or appropriate conversational style (e.g., encouraging, to the point, etc.), and administrator-defined case-specific attributes (e.g., competence across a set of detailed skills that the user is seeking to master).

[0069] In some embodiments, the personalization approach can be adjusted over time by using a recurrent neural network (RNN), such as a long-short-term memory network (LSTM), to model the temporal nature of user interactions and the relationship between past interaction patterns and future behavior. The resulting trained model is then used to optimize the appropriate level of active user engagement.

[0070] In some embodiments, the system may be configured to model one or more aspects of biological memory, such as episodic memory of user interactions and / or semantic memory of user preferences. In some embodiments, cognitive science models of knowledge representation may be used to structure user profiles. For example, learning science principles such as desired difficulty, power law decay of episodic memory, spaced repetition, and retrieval practice may be used to strengthen memories important to the learner and / or estimate the user's memory for past interactions. In some embodiments, other principles employed include temporal reframing or embodying self or others via conversational AI to help users shift perspectives. For example, a user may simulate a conversation with their future or past self to help them make a decision, or an advisor or consultant may simulate a conversation.

[0071] In some embodiments, a profile may store not only explicit attributes such as demographics, context, and activity logs, but also semantic embeddings obtained from conversational data, and in particular insights about the user gained from interactions. These interactions may be direct (e.g., explicitly provided by the user to improve the user's experience), indirect and active (e.g., a conversational agent may prompt the user with questions or interactions designed to elicit useful information for the user profile), or indirect and passive (e.g., a conversational agent infers attributes from interaction history or other data provided about the user, or from other visual or mechanical user interactions such as text, clicks, dwell times, etc.).

[0072] In some embodiments, the user's needs, motivations, and / or other factors may be inferred, for example, by a profile update module. In some embodiments, the profile update module uses one or more calls to an LLM or other updatable reinforcement learning model to input the raw data described above, extract information related to the user profile, and pass this information to a vector encoder, which may store it as a vector encoding or in another suitable format. In some embodiments, the user profile may be represented as a flexible, hierarchical object, such as a json, xml, or dictionary object containing key-value pairs, where the key describes an attribute and the value defines the current state of that attribute for that user. The user profile representation may be viewable and editable by the user, allowing the user control over both the user's experience with the conversational agent and the information about the interaction that may be saved.

[0073] In various embodiments, the short-term, near-term, and long-term adaptation processes can focus on updating different types of user data and profiles. Daily user activity patterns can update near-term interests. Data from a lifetime of interactions can shape long-term motivations and needs. The relative influence of old and new data can be controlled by a parameterized age-dependent decay function. This decay function can vary across subjects, allowing, for example, the significance of information about recent grocery store purchases to decay more quickly than the purchase of a new car.

[0074] In some embodiments, user interactions may be further personalized by referencing a long-term personal conversational memory specific to each user. Past conversations may be stored as a message history, for example, as an ordered list of messages and responses between the user and an agent. These conversation histories may be embedded in the same vector space, using techniques similar to the retrieval-augmented generation methods described herein with respect to the brand learning module 110 (but with differences such as varying chunk size to match message length). In some embodiments, the conversation history is searched at runtime, as described herein, and relevant information is inserted into system prompts at runtime. Conversations may be filtered for each user, so that embedded conversation snippets may only be retrieved in the context of new conversations involving the same user.

[0075] In some embodiments, a conversation memory tool (similar to the codex tool described herein) can be used to control the circumstances under which the conversation memory should be queried. For example, when a customer asks a customer service conversation agent a question, the entirety of past conversations with the user can be queried to identify related problems, such as the series of steps the customer has already attempted in the past to solve the problem. In this way, the conversation agent can avoid repeating the same advice and instead provide more useful assistance that is modified based on this new context (i.e., the steps the user has already attempted).

[0076] In some embodiments, conversation history may be filtered or summarized before embedding to optimize storage, enhance search, or increase privacy. For example, conversation history may be turned off or limited by a user based on user-controlled settings. In some embodiments, conversation history may be stored hierarchically. For example, an entire conversation (a set of messages within a time frame, such as the past hour, or an entire conversation about a set of related topics) may be summarized and embedded instead of or in addition to the messages that make up the conversation. In this way, long histories can be efficiently searched by referencing the conversation summary, and entire related conversations, rather than just snippets and individual messages, may be searched for and inserted into system prompts.

[0077] In some embodiments, the user experience may be further personalized and tailored in real time, for example, by a semantic router, which is a module that uses LLM or other models to identify the needs, preferences, intent, and other key characteristics of a user and their request in order to route the user request to the most appropriate conversational agent among multiple candidate conversational agents.

[0078] In some example embodiments, the adaptive curation module 130 can leverage information from the brand learning module 110 and / or personalized rapport module 120 to generate customized content recommendations for each individual user. Example embodiments of the process or model can access a smart recommendation database that can contain a wide range of recommendations for products, services, study plans, etc. In some embodiments, this database can be populated by an administrator and used in combination with evolving user profiles and machine learning algorithms to make these smart recommendations. In other examples, the information in the database can be structured data or generated by processing and categorizing unstructured data.

[0079] In some example embodiments, the personalized rapport module 120 provides information about the user's interactions with the system, such as the user's preferences and behavior, which is used in combination with data from the brand learning module 110 to form a comprehensive understanding of the user. In other words, the personalized rapport model can process feedback information corresponding to the user or their interactions with the system. The feedback data can include one or more of explicit feedback and implicit feedback. This information can be used by the adaptive curation module 130 to generate personalized content recommendations tailored to each user's individual needs and preferences.

[0080] In some example embodiments, the adaptive curation module 130 continuously updates these recommendations based on user interactions, such as whether the user accepts a recommendation, and uses this information to refine subsequent recommendations. In this way, the module can provide a highly customized and personalized experience for individual users, ensuring that recommendations are accurate, meaningful, and relevant to the user's needs.

[0081] Additionally, in some exemplary embodiments, the adaptive curation module 130 can continuously update one or more recommendations based on various factors, including, but not limited to, user interactions with the system and other data points indicative of recommendation acceptance. This information can be used to refine subsequent recommendations, taking into account not only whether a recommendation was accepted, but also the timing and degree of acceptance. In some embodiments, this enables the module to better infer user preferences and needs and provide more personalized and relevant recommendations over time. For example, the module can determine one or more scores corresponding to user characteristics (e.g., based on model parameters learned during training based on recordings indicative of user characteristics, AI responses, and user feedback on those responses) that can be used to generate AI responses tailored to the user characteristics.

[0082] In some embodiments, the adaptive curation module 130 may use a feed-forward neural network to match user preferences with item attributes and generate recommendations over time, such as:

[0083] The neural network may be trained based on a set of user-item interaction data, including a user profile (as described herein). The user profile may include demographic data, personality traits, and / or past engagement data with items, which may be associated with item metadata attributes, i.e., text descriptions, audio and visual features, popularity metrics, and / or embedded category vectors. Notably, in some embodiments, the user profile data may include not only the conversation history but also current conversation attributes, such as the user's current goals, mood, and other factors that an emotionally intelligent and socially competent person might notice during the course of a sales conversation, as inferred by the semantic router described above.

[0084] In some embodiments, item attributes can be extracted from item metadata by an attribute encoding layer of a neural network that generates a multidimensional item attribute vector. The item metadata includes descriptions of the types of users for whom an item would be valuable and user attributes, etc. In some embodiments, a scoring and ranking layer of the neural network can receive the user preference vector and the item attribute vector as input and calculate a relevance score between each user-item pair, for example, by distance calculation techniques in a joint embedding space or other relevance scoring techniques. Confidence values ​​for recommendations are calculated from the relevance scores, for example, using a sigmoid function transform, and a threshold is used to filter out low-confidence recommendations.

[0085] In some embodiments, training data for this predictive feed-forward neural network model, such as an RNN or LSTM, can be obtained in many ways, such as by observing transcripts of past conversations between customers and human salespeople in sufficiently similar contexts and combining them with the sales outcomes after those conversations. New training data can be continuously generated within the system, for example, by storing anonymized conversation history and outcome pairs, and further refined by varying the conversational agent's approach from conversation to conversation to more effectively explore the conversation-outcome pair space.

[0086] In some embodiments, the adaptive curation module 130 can dynamically refine the user preference encoding and item attribute encoding neural network layers based on collected recommendation feedback data indicating user behavior toward recommended items, in addition to full retraining based on datasets obtained using the methods described herein. Positive interactions, such as clicks, purchases, or positive emotional responses to recommended items, can trigger incremental adjustments in the preference encoding layers to strengthen preference signals for the associated item attributes, while negative interactions can trigger incremental adjustments to weaken preference encodings for those item attributes. In some embodiments, adjustments can be made proportional to the recommendation confidence calculated at the time of recommendation, such that recommendations with higher confidence have a greater training effect.

[0087] In some embodiments, other approaches to predictive translation can be applied in addition to or instead of a neural network model. For example, in some embodiments, an explicit intent analysis approach may be implemented by a processor as described herein. In some embodiments, a user's most recent messages are chunked and embedded into a vector space as described herein, and then compared to a set of trigger vectors in the embedding space. The trigger vectors may be embedded representations of message sequences previously shown to precede a particular purchase or decision (e.g., based on past user conversations that led to a particular outcome), or they may be handwritten canonical messages that an administrator reasonably estimates may precede such behavior. For example, an administrator may add trigger vectors to the embedding of one or more descriptions of problems that the administrator's product would be a good solution to. If a new user's message is embedded and located sufficiently close to one or more of these trigger vectors, as measured, for example, by cosine distance, the conversational agent can share details about the product and how it solves the user's problem.

[0088] In some embodiments, predictive transformation is not limited to direct business transactions, but may be applied more generally to prompt users to take action or progress toward goals at appropriate times, such as in educational applications that prompt users to complete educationally beneficial tasks or actions, or in personal growth applications that prompt users to examine and / or adjust their habits in ways that will be beneficial in the long term.

[0089] In some example embodiments, the intelligence and insights module 140 can analyze inputs from various sources to extract useful insights and provide data-driven recommendations. This analysis can be performed using advanced machine learning algorithms, such as deep learning and predictive analytics. These algorithms process large amounts of data to identify patterns and trends in user behavior and preferences, allowing the system to better understand the motivations and needs of individual users. In some embodiments, the intelligence and insights module 140 can provide one or more data-driven recommendations, such as for improving the responses of respective conversational agents, improving one or more services offered, and / or improving system performance.

[0090] Examples of user behavioral and preference patterns and trends may include your preferred communication style, purchasing habits, types of products or services you are interested in, your level of engagement with various content or media, etc. This information is used to provide you with more personalized and relevant content and recommendations, improving your overall user experience.

[0091] In some exemplary embodiments, the intelligence and insights module can analyze inputs from various sources to gain a comprehensive understanding of both the performance of the system and the needs and preferences of users. Some examples of insights generated about the performance of the system include identifying areas where the system can be optimized to improve user engagement and satisfaction, or determining which modules or components are performing well and which may require further improvement.

[0092] In some exemplary embodiments, a score corresponding to a user's needs and preferences may be inferred. For example, the module may infer insights such as identifying what types of content are most appealing to a particular user or identifying which products or services they may be interested in based on the user's behavior and preferences (e.g., scores based on obtained feedback).

[0093] In some exemplary embodiments, past interaction history, needs, goals, preferences, or other information about a user, such as those described in the preceding paragraphs, can be presented to the user directly, e.g., in a human-readable and understandable format, so that relevance and accuracy can be more directly assessed. For example, the user can view, delete, modify, or add to the memories the system has about them. This allows the system to be more accurately tailored to the user's actual current state and to build stronger trust with the user through greater transparency, as opposed to less transparent personal data tracking in the services of, for example, online advertising networks or social media platforms.

[0094] In some example embodiments, predictions made by the module may be based on analysis of data collected from various inputs. Machine learning algorithms can be used to identify patterns and trends in user behavior and preferences, and this information is used to make predictions about which products or services a particular user may be interested in. For example, one or more models can be trained to output a store that exhibits various patterns, behaviors, or preferences corresponding to the user. In some embodiments, predictions are continually refined over time based on the user's interactions with the system, allowing the module to provide increasingly accurate and relevant recommendations to individual users.

[0095] In some example embodiments, the intelligence and insights module 140 can use data-driven insights and advanced machine learning algorithms to generate analytics about the performance of the system and the effectiveness of each module. For example, in some embodiments, one or more models can be trained to evaluate the effectiveness of the AI ​​system and score changes to the system based on whether the changed outputs result in more accurate or improved results (e.g., based on feedback or feedback scores indicating those improvements). This information is used to optimize and continuously improve the system, ensuring it remains at the cutting edge of conversational AI technology. The module can provide insights into the performance of various components and how they are impacting the user experience.

[0096] In some exemplary embodiments, the output from one or more modules is not simply a traditional report, but leverages conversational AI techniques to enable managers to naturally converse about the insights and gain a deeper understanding of the data. The insights are generated based on an analysis of inputs from various sources, such as user interactions, a brand learning module, a personalized rapport module, and an adaptive curation module.

[0097] In some embodiments, users can learn characteristics that correspond to different individual users. Embodiments may provide these characteristics as inputs to a model, in conjunction with other inputs, and adjust the model's weights or biases based on the characteristics. In some embodiments, the characteristics may correspond to parameters of one or more models trained on a particular user or a set of users determined to have similar preferences. In some exemplary embodiments, an AI system can use a combination of machine learning algorithms, such as natural language processing (NLP), to analyze user behavior and preferences and generate personalized recommendations. The output can be a single optimal formula that applies to all users, or it can be trained for smaller subsets of similar users, or even for individual users based on each user's specific needs and preferences. The system can continuously update and evolve the formula (or its parameter weights and biases) based on one or more user interactions, ensuring that recommendations remain relevant and personalized over time.

[0098] In some embodiments, intelligence and insights module 140 may be configured to ingest and process multiple inputs, such as user-agent conversation logs including dialog history with full transcripts, user profile data attributes including interests, preferences, purchase history, and other derived attributes, interaction and engagement metrics by agent / user / topic area / question type / etc. segmentation (including any segmentation performed by an administrator during analysis), and / or product catalog metadata defining available items, topics, and intent.

[0099] In some embodiments, given storage constraints, conversation logs may be sampled to extract a representative dataset. Conversations may be embedded in a manner described herein with respect to the personalized rapport module 120, such as by message embedding and / or conversation summaries, to form a hierarchical dataset. Conversations may be further grouped and summarized by time, subject matter, user characteristics, or other attributes, and hierarchical levels may be added above the conversation level where conversations are summarized by group. In some embodiments, this hierarchical structure allows administrators to ask very broad questions about broad conversations and quickly receive analysis based on the LLM analysis, while also allowing them to drill down further into individual conversations with more specific queries. For example, vector search is enabled in the manner described herein for knowledge base datasets related to the brand learning module 110.

[0100] In some embodiments, topic modeling may be performed at one or more layers of this hierarchical dataset, such as at a higher-level summarization layer above the conversation layer, to identify topics and / or aggregate user needs. In some embodiments, a recommender system may match profile vectors with conversational features to suggest additions to the knowledge base or identify groups of users who are most likely to benefit from a particular type of interaction, such as proactively suggesting exercises to students with a particular misconception that have been shown to be effective in reducing that misconception in other students.

[0101] In some embodiments, the management module 170 (FIG. 1) can be configured to display, via an interactive management UI, for example, detailed, concise, and easy-to-understand topic summaries generated by the LLM and tailored to highlight important or noteworthy trends; sample conversations for qualitative evaluation; charts of topic trends, engagement, and user needs over time; recommendations for prioritized knowledge gaps that limit agent effectiveness (e.g., achieved by categorizing user responses into satisfactory / unsatisfied clusters and identifying the most common user needs from conversations that ended with user messages classified as “unsatisfied”); user preference clusters segmented to identify, for example, underserved audiences; and the like. In some embodiments, the management UI can be configured to provide a natural language interface with which an administrator can interact. For example, an administrator can explore insights through natural language conversations with the system by asking follow-up questions and requesting additional details in the manner described above with respect to higher-level conversation summaries.

[0102] Various embodiments of the enhanced conversational AI system 100 may include different or other components than those illustrated herein, such as various databases or other data storage components that may contain various records and data structures corresponding to the data flow between different components of the system. Furthermore, the databases may store various training data that may include training and validation records. The training datasets may be augmented over time with additional records as the AI ​​system operates to improve the performance of one or more models trained on a subset of one or more records included in the training dataset. For example, feedback data obtained related to model inputs or outputs may be used to generate one or more records for training to improve model performance.

[0103] Additional / alternative components of an AI system may include, but are not limited to, the following: (1) LLM / Natural Language Processing (NLP) module: This module analyzes and processes human language input. It utilizes advanced NLP techniques such as sentiment analysis, named entity recognition, and text classification to understand and respond to user input in a human-like manner. (2) Knowledge Base: A central repository for storing and organizing the data, information, and knowledge acquired by an AI system. It may employ advanced techniques in data management, such as embedding and indexing, to make this information easily accessible and usable by other components of the system. (3) User profile databases, which may store data related to individual users, such as preferences, interaction history, personal information, and processed or abstracted forms of this data. This data is collected and updated through users' interactions with AI systems and may be used to provide each user with a highly customized and personalized experience. (4) Interaction Interface (User Interface), which provides a means for users to interact with the AI ​​system. This may be a conversational agent, chatbot, virtual assistant, or other conversational interface, or a flexible search box that allows queries or other prompts, or the upload of images, videos, audio, data files, or other formats, or a combination of the above, providing a simple and intuitive way for users to engage with the AI ​​system. (5) Various machine learning algorithms, including training various machine learning models, may be used to analyze the data described herein. They may be employed by one or more components or modules described herein to make predictions, provide insights and recommendations, etc. These algorithms may utilize techniques such as deep learning, reinforcement learning, and predictive analytics, and are constantly learning and evolving to provide more accurate and meaningful insights and recommendations over time. (6) Various internal or external data sources that an AI system analyzes or incorporates into its training data, such as data inputs, articles, videos, podcasts, and other forms of multimedia content. These inputs are processed and analyzed by machine learning algorithms to gain a deeper understanding of user preferences, behaviors, and needs.

[0104] 2 illustrates an example method 200 for providing an adaptive AI-driven conversational agent, according to at least one embodiment. Various embodiments may implement an AI-driven system, such as the personalization-enhanced conversational AI system 100 (described in detail herein). In some embodiments, the method 200 may be executed on a computer having a processor, a memory, and one or more sets of code stored in the memory and executed by the processor. When executed, the sets of code configure the processor to perform the steps of the method 200 described herein.

[0105] In some embodiments, method 200 may begin at step 210, in which a processor is configured to ingest a first branded content dataset. As described herein, in some embodiments, the processor may implement a dedicated module, such as a content ingestion module, that ingests proprietary and / or other content from various sources, such as articles, videos, podcasts, policy documents, technical proposals, interview recordings, and social media recordings, for use in training the AI. The content is processed to extract relevant features and stored in a content database. The processed content is passed to an embedding model and stored as a set of indexed vectors that can later be used by the system for retrieval-augmented generation. This allows the user interface to quickly access this information as relevant context at runtime. In some embodiments, ingesting the first branded content dataset may include processing the first branded content dataset with one or more machine learning algorithms, as described herein, to identify one or more insights about the first branded content dataset. In some embodiments, the insights may include tone, language, and at least one of audience engagement, intent, mood, receptivity, skill, expertise, and / or understanding, among others.

[0106] In step 220, in some embodiments, the processor is configured to organize the ingested first brand content dataset into multiple embeddings and indexes and store the multiple embeddings and indexes in a knowledge base. As described herein, in some embodiments, the processor can implement a dedicated module, such as brand learning module 110, which uses ingested brand content, such as articles, videos, and podcasts, to train an AI to understand the essence and DNA of the brand. This module uses natural language processing (NLP) algorithms to analyze the brand content, focusing on aspects such as tone, language, and audience engagement, to gain a deep understanding of the brand's unique identity. The analysis results are stored in the knowledge base, which serves as a comprehensive source of information about the brand. The AI ​​continuously updates its understanding of the brand as new content is ingested and processed, resulting in a set of learned representations of the brand that other modules reference to deliver highly customized and personalized user experiences.

[0107] In some embodiments, the embedding may include embedding of a vector within an embedding space. In some embodiments, the location of the embedding may provide a semantic meaning for the content represented by the vector. In some embodiments, the index may include a data structure that provides a mapping between branded content data and its location within the knowledge base, as well as links to metadata associated with the content data. In some embodiments, the embedding space may include a mathematical vector space that captures semantic relationships between branded content, and embedding branded content as indexed vectors in this space enables identification of contextual similarities between content.

[0108] In step 230, in some embodiments, the processor may be configured to generate a user profile for each user based at least in part on the organized plurality of embeddings and indexes and the user history for each user associated with each user profile. In some embodiments, the user profile may be accessed by the user via a user interface, for example, on a user device of end user 150. In some embodiments, the user interface may be implemented by a user interface module that is responsible for interacting with the user and collecting data regarding the user's preferences and needs. The interface may be in the form of a chatbot, a voice-based system, or any other suitable interface. The collected data may be stored in a dedicated user database.

[0109] In step 240, the processor is configured to update the first user profile based at least in part on one or more interactions between the first user and the first user profile and one or more models trained on recordings indicative of one or more processes of the human user. In some embodiments, the one or more models trained on records indicative of one or more processes of the human user include at least one or more models trained on records indicative of biological-cognitive and neuroscience processes of the human user that provide information about at least one of the user's thought processes, behavioral patterns, motivations, or biases. In some embodiments, the one or more models include neuro-linguistic processes and / or reinforcement learning algorithms trained on user query and response pairs to optimize the system's response.

[0110] In some embodiments, the processor may be configured to implement a personalized rapport module. This module uses user data collected from the interface to generate a personalized experience for the user. This AI leverages advances in cognitive science and neuroscience to proactively engage the user and create lasting connections. The output of this module may be a personalized set of engagement strategies for the user, as described herein, that may influence subsequent responses. In some embodiments, the first user profile may be updated, for example, continuously in real time or periodically, based on a stream of user interaction data to ensure accuracy and personalization of the system output.

[0111] At step 250, the processor is configured to personalize one or more responses of the conversational agent interacting with the first user based at least in part on the first user profile, which may include, for example, responses based on dedicated and / or curated memory, information, etc., as described herein. In some embodiments, the processor may interact with the user via a user interface.

[0112] At step 260, the processor is configured to generate and provide to the first user, via a conversational agent, customized content recommendations for the first user based at least in part on the first user profile. In some embodiments, the processor may implement an adaptive curation module that provides individualized content across all media and platforms based on user data and a personalized rapport strategy. The content may include blog posts, images, and other media. The module may also be capable of making recommendations based on machine learning analysis of user interactions.

[0113] In step 270, inputs from multiple users responding to interactions with the respective conversational agents are analyzed to extract insights related to the interactions with the multiple users and provide data-driven recommendations. In some embodiments, the processor may implement an intelligence and insights module that provides reports and insights regarding user behavior and trends. Generative AI is used to analyze all user conversations and provide insights and recommendations to the administrator. The output of this module may be a set of actionable insights and recommendations for the administrator regarding content creation, etc.

[0114] The above-illustrated flow of methods (and / or computer program instructions) may be implemented within an AI system and may include other examples of modules and corresponding functions described herein. In other embodiments, the illustrated processing may be distributed across fewer, other, or different components. These modules and databases interact with each other to form a complete AI-powered system for engaging with users in a personalized and adaptive way.

[0115] In some embodiments, a machine learning model (model) described herein can take in one or more inputs and generate one or more outputs. Examples of machine learning models can include neural networks or other machine learning models described herein, which can take in inputs (e.g., the example input data described above) and provide outputs (e.g., output data as described above) based on the inputs and parameter values ​​of the model. For example, a model can be provided with an input or set of inputs to process and provide an output or set of outputs based on user feedback data or outputs determined by another model. In some cases, the outputs can be fed back to the machine learning model (e.g., alone or in combination with an indicator of performance of the output, a threshold associated with the input, or other feedback information) as inputs to train the machine learning model. In some embodiments, a machine learning model can update its configuration (e.g., weights, biases, or other parameters) based on the results of evaluating a prediction or instruction (e.g., output) against feedback information (e.g., scores, rankings, text responses, or other feedback information) or the outputs of other models (e.g., scores, rankings, user characteristics, etc.). In some embodiments, such as when the machine learning model is a neural network, connection weights may be adjusted to accommodate discrepancies between the neural network's predictions or instructions and feedback data. In some embodiments, one or more neurons (or nodes) of the neural network may require their errors to be sent back to them through the neural network to facilitate the update process (e.g., backpropagation). Updates to connection weights may reflect, for example, the magnitude of the error to be propagated backward after forward propagation is complete. In this manner, the machine learning model may be trained to, for example, improve the predictions or instructions it generates.

[0116] In additional embodiments, users are provided with transparency into memory retention policies between collaborating agents, with explicit visibility into data usage, sharing protocols, retention period specifications, and the option to purge memory on demand through an administrator dashboard.

[0117] In certain embodiments, users have fine-grained control over granting and revoking multi-agent memory access over time, on a per-agent or per-memory group basis. For example, trip-related memory may be granted access to a trip planning agent for a limited time, with access limited to the context of the upcoming trip, and then automatically revoked after the trip to maintain privacy.

[0118] In some embodiments, the machine learning model may include an artificial neural network. In such embodiments, the machine learning model may include an input layer and one or more hidden layers. Each neural unit of the machine learning model may be connected to one or more other neural units of the machine learning model. Such connections may have a constructive or inhibitory effect on the activation state of the connected neural units. Each neural unit may have a summation function that combines the values ​​of one or more of its inputs. Each connection (or the neural unit itself) may have a threshold function that a signal must exceed before being propagated to other neural units. Machine learning models may learn and train automatically rather than being explicitly programmed, and may significantly outperform computer programs that do not utilize machine learning in solving problems in a given domain. During training, the output layer of the machine learning model may correspond to a classification, and inputs known to correspond to the classification may be input to the input layer of the machine learning model during training. During testing, inputs with unknown classifications may be input to the input layer, and a classification may be determined and output. In some embodiments, the classification may be an indication of whether the natural language text is predicted to optimize an objective function that satisfies a user's preferences, or an indication of whether the natural language text (or text) provided by a user corresponds to a classification of attributes or characteristics predicted to correspond to that user. In some embodiments, the classification may be an indication of a user's characteristics determined from the natural language text, such as based on a vector representing the natural language text, or an indication of whether a vector representing the generated natural language text is predicted to match the user's preferences (which may be determined based on the user's characteristics). In some embodiments, the classification may be an indication of the embedding of the vector within an embedding space of the natural language text represented by the vector. In some embodiments, different regions in the embedding space may correspond to different ways in which text responses are generated, such as based on the user's inferred preferences.Some example machine learning models may include one or more embedding layers in which information or data (e.g., any of the data or information discussed herein in connection with the example models) is converted into one or more vector representations, which may be pooled in one or more subsequent layers to convert the one or more vector representations into a single vector representation.

[0119] In some embodiments, the machine learning model may be configured as a factorization machine model. The machine learning model is a nonlinear or supervised learning model capable of performing classification or regression. For example, the machine learning model may be a general-purpose supervised learning algorithm that the system uses for both classification and regression tasks. Alternatively, the machine learning model may include a Bayesian model configured to perform variational inference (e.g., deviation or convergence) of an input from previously processed data (or other inputs in a set of inputs). The machine learning model may be implemented as a decision tree or as an ensemble model (e.g., using random forests, bagging, adaptive boosters, gradient boosting, XGBoost, etc.). In some embodiments, the machine learning model may incorporate one or more linear models in which one or more features are preprocessed or the output is postprocessed, and training the model may include training with or without preprocessing or postprocessing with such models.

[0120] In some embodiments, the machine learning model implements deep learning via one or more neural networks (one or more of which may be recurrent neural networks). For example, some embodiments may reduce the dimensionality of high-dimensional data (e.g., having one million or more dimensions) before providing it to the learning model, such as by forming a latent space embedding vector based on the high-dimensional data (e.g., natural language text) to reduce processing complexity, as described in various embodiments herein. In some embodiments, the high-dimensional data may be reduced by an encoder model (which may implement a neural network) that processes vectors or other data output by an NLP model. For example, training a machine learning model may include generating multiple latent space embeddings as or in association with the output of a classified model. Each of the multiple models discussed herein may determine or perform processing based on the spatial embedding and known latent space embeddings, as well as the distance between the embeddings. Alternatively, a score may be determined that indicates whether a user's preferences are represented by one or more embeddings or regions of multiple embeddings, such as when generating an AI response based on the user's learned preferences.

[0121] Examples of machine learning models include multiple models. For example, a clustering model can cluster latent space embeddings represented in training (or output) data. In some cases, the ranking or other classification of one or more latent space embeddings within a cluster can reveal information about other latent space embeddings within or assigned to the cluster. For example, a clustering model (e.g., K-means, density-based spatial clustering of applications with noise (DBSCAN), or various other unsupervised machine learning models used for clustering) can take a latent space embedding as input and determine whether it belongs to one or more other clusters of other previously trained spatial embeddings (e.g., based on a threshold distance). In some embodiments, a representative embedding for a cluster containing multiple embeddings can be determined, e.g., by obtaining a ranking through one or more samplings of the cluster and selecting a representative embedding based on the ranking. The representative embedding can then be sampled (e.g., more frequently) for ranking purposes relative to other embeddings outside the cluster or representative embeddings of other clusters. This allows the learning process to determine whether a new user is similar to one or more other users and bootstrap response generation based on inferred preferences for the new user based on a reduced set of known characteristics similar to those of the other users.

[0122] In some exemplary embodiments, an AI system employing one or more of the present technologies may incorporate additional modules to enhance the overall functionality and performance of the system. For example, a sentiment analysis module may be incorporated to better understand a user's emotions and reactions to content. This module analyzes the user's tone and phrasing, as well as facial expressions and other factors such as body language, to determine the user's emotional state. This information is used to further improve the personalized rapport and adaptive curation modules by gaining a more comprehensive understanding of the user's preferences and needs.

[0123] In some exemplary embodiments, an AI system employing one or more of the present techniques may incorporate a multilingual support module to enable the system to interact with users in multiple languages, thereby providing a more inclusive user experience by allowing users to engage with the system in their preferred language.

[0124] In some exemplary embodiments, an AI system employing one or more of the present techniques may incorporate a data privacy module to ensure that user data is securely stored and managed in accordance with privacy regulations. This module oversees the storage and handling of user data, protects the data from unauthorized access and breaches, and ensures that it is managed in a manner compliant with relevant privacy laws and regulations. For example, an AI system used by a healthcare institution may store data in compliance with HIPAA regulations.

[0125] In some exemplary embodiments, an AI system employing one or more of the present technologies can be incorporated into a robot that interacts with a user in real time. The robot may be configured with a conversational interface that includes speech-to-text capabilities to understand a user's voice input and text-to-speech capabilities to respond to the user in a natural and intuitive manner. Furthermore, the interface may be multilingual, making it accessible to a wider range of users. The robot uses machine learning algorithms to understand user preferences, provide tailored content, and make smart recommendations. The robot also stores user data and uses it to continuously improve user interactions. Insights generated from conversations can be fed back to a management interface to inform decisions regarding content creation, user behavior trends, and product recommendations.

[0126] In some exemplary embodiments, an AI system employing one or more of the present technologies may be a mobile application that utilizes location-based data and voice to improve interaction with users on the move. This variation integrates four modules—a brand learning module, a personalized rapport module, an adaptive curation module, and an intelligence and insights module—with location data, enabling the system to provide content and recommendations that are optimally tailored to the user's physical location. The mobile application includes sensors such as a GPS and accelerometer to collect location data and a microphone to collect voice input from the user. An audio interface, such as earphones, enables real-time, on-the-go interaction between the user and the AI ​​system, further enhancing the user experience by incorporating location data and voice input into the decision-making process. Additionally, this embodiment may be designed to interface with other wearable devices or sensors, such as heart rate monitors or fitness trackers, to collect additional data and provide more context for the system to make more informed recommendations.

[0127] In some exemplary embodiments, an AI system employing one or more of the present technologies may be a virtual reality platform incorporating four modules: a brand learning module, a personalized rapport module, an adaptive curation module, and an intelligence and insights module. This embodiment uses advanced sensing technologies, such as haptic feedback and eye tracking, to provide users with a highly immersive and interactive virtual experience. The system can dynamically curate content and make recommendations based on the user's real-time reactions, behaviors, and preferences within the virtual environment. For example, if a user's engagement with a certain type of content increases, the system can recommend similar content to further enhance the user's virtual experience. The system can also utilize biometric data, such as heart rate and brain activity, to make more informed decisions. For example, if a user's heart rate increases while viewing a certain type of content, the system can recommend different or similar content to help the user relax and immerse themselves in the virtual environment. This embodiment has the potential to revolutionize how people interact with technology, media, and information in a highly engaging and personalized way.

[0128] In some example embodiments, an AI system employing one or more of the present technologies may include a connection between the AI ​​system and an external learning management system (LMS) or student record system. This variation allows for the integration of user data from the LMS or student record system into four modules: a brand learning module, a personalized rapport module, an adaptive curation module, and an intelligence and insights module. This integration works by transferring the necessary data from the LMS or student record system to the AI ​​system using technical means such as an API. This integration enables the AI ​​system to provide individually tailored content and recommendations based on the user's learning history, educational background, detailed proficiency in relevant skills, and academic goals. The AI ​​system uses this data to create a personalized learning plan for each user, improving the efficiency and effectiveness of the educational experience. The connection between the AI ​​system and the LMS or student record system allows for real-time updates and data sharing, further improving the user experience.

[0129] In some example embodiments, an AI system employing one or more of the present techniques may be integrated with blockchain technology, which leverages distributed data storage and secure cryptographic protocols. This integration provides a secure, tamper-proof solution for storing user data and interactions, ensuring data privacy and protection. In this embodiment, the AI ​​system is designed to interact with smart contracts, enabling automated decision-making and improving the speed and efficiency of data processing, with safeguards in place to prevent the AI ​​system from making irreversible transactions on the blockchain without proper authorization. In addition to providing secure data storage, blockchain technology also enables the creation of unique digital tokens that users can earn and trade based on their interactions with the AI ​​system. This could be a new way to incentivize user engagement and create more immersive experiences. The decentralized nature of blockchain ensures transparency and accountability in the tracking and distribution of these tokens, further improving the user experience.

[0130] In some exemplary embodiments, an AI system employing one or more of the present technologies may be a micropayment-enabled system that utilizes machine learning algorithms to optimize user interactions. This variation integrates a micropayment module with four modules: a brand learning module, a personalized rapport module, an adaptive curation module, and an intelligence and insights module. The micropayment module allows for real-time testing and optimization of various content and recommendation candidates. The system uses A / B testing to determine the most effective candidates for individual users and uses that data to determine which content to provide and when. Additionally, the micropayment module allows users to make small payments for access to premium content or additional features, providing a new revenue stream for the system. This embodiment may be designed to interface with other systems, such as a learning management system or student record system, to collect additional data and provide more context for the system to make recommendations.

[0131] In some example embodiments, an AI system employing one or more of the present technologies may include a payment facilitator system that integrates four modules—a brand learning module, a personalized rapport module, an adaptive curation module, and an intelligence and insights module—with a third-party payment gateway. In this embodiment, the system provides users with the option to securely pay using various payment methods, such as credit cards, digital wallets, and bank transfers. The system is configured to use machine learning algorithms to optimize payment processing and ensure a seamless user experience, and further integrates with a third-party payment gateway to offer a full range of payment options. This can bring new revenue streams to the system and provide users with a convenient and secure means to access premium content and features.

[0132] In some example embodiments, an AI system employing one or more of the present technologies may incorporate a personalized pricing model into the conversational AI system. In this variation, a variety of factors, such as user behavior, engagement, and other data, are considered to dynamically determine an appropriate price for each user. The conversational AI system can then make subscription or micropayment recommendations based on this personalized pricing model, thereby providing a more tailored and engaging experience for users. The system continuously updates pricing in real time based on changes in user behavior, ensuring users always receive the most relevant and accurate pricing information. Additionally, the present embodiments may be integrated with blockchain technology, providing a secure and tamper-proof solution for storing and processing payment transactions.

[0133] In some example embodiments, each conversational agent is assigned specialized roles and capabilities while sharing access to individual user profiles and memories. For example, personal health, education, and travel assistant agents may be tasked with maintaining memories specific to each topic. When accessed by a user, the agents can coordinate the exchange of memory vectors, enabling seamless, personalized handoffs between conversations.

[0134] In additional embodiments, conversational agents can have enhanced memory monitoring and triggering capabilities. The agents can actively track updates to a user's memory profile and react autonomously based on pre-defined triggers. For example, a health agent can initiate a new conversation with dietary advice each time a user adds a diagnostic memory that meets predetermined conditions. Similarly, an education agent can initiate a progress check or supplemental resources each time an updated learning memory indicates the user has completed an associated training course or template.

[0135] In some exemplary embodiments, an AI system employing one or more of the present technologies can be integrated into an existing product, such as a website or application, to provide core conversational AI functionality within the existing technology. This integration enables a seamless user experience by integrating chat functionality into the existing product's familiar interface. This embodiment leverages the power of the brand learning module, personalized rapport module, adaptive curation module, and intelligence and insights module to provide users with a customized and personalized experience within the existing product. By leveraging machine learning algorithms, the system can continuously analyze user interactions and generate insights to improve the user experience. This embodiment offers the advantage of combining the benefits of conversational AI technology with the existing product's familiar interface, providing users with a seamless and personalized experience.

[0136] Three non-limiting use cases are presented: (1) celebrity conversation AI, (2) expert conversation AI (health and nutrition), and (3) business conversation AI (schools). In at least some use cases of AI systems, end users can interact with the system through a consumer-facing interface, such as a website or a chat feature built into their phone. The AI ​​system can obtain or infer input from the user, such as the user's preferences and conversation history, to tailor the interaction to the user's unique needs and interests.

[0137] In some embodiments, an administrator interface allows administrators to access the intelligence and insights generated by the conversational AI system. Example user interface views include data visualizations of end-user activity, summaries of trending topics, and the ability to converse with the system to access insights, recommendations, and predictions based on the full dataset of end-user interactions. This provides a powerful tool for informed decision-making and system optimization.

[0138] 3 is an example user interface implementing an administration (admin) interface in accordance with at least one embodiment. This module provides an interface for administrators to access insights and recommendations generated by the intelligence and insights module 140. Administrators can use this interface to monitor user behavior, make predictions about user behavior, and make decisions regarding content creation and other related topics. In various embodiments, the admin screen 300 illustrates the use of the intent and memory tracking features enabled.

[0139] 4 is an example of a user interface implementing a conversational agent, according to at least one embodiment. As shown in user interface 400, the conversational agent accurately remembers the conversation and provides advice according to the specific content associated with the user's user profile.

[0140] 5 is a physical architecture block diagram illustrating an example of a computing device (or data processing system) that may implement aspects of the technology described above. Various portions of the systems and methods described herein may include or be executed on one or more computer systems similar to computer system 1000. Furthermore, the processes and modules or subsystems described herein may be executed by one or more processing systems similar to that of computer system 1000.

[0141] The computer system 1000 may include one or more processors (e.g., processors 1010a-1010n) coupled to a system memory 1020, an input / output (I / O) device interface 1030, and a network interface 1040 via an input / output (I / O) interface 1050. The processor may include a single processor or multiple processors (e.g., distributed processors). The processor may be any suitable processor capable of executing instructions. The processor may include a central processing unit (CPU) that executes program instructions to perform arithmetic, logical, and input / output operations for the computer system 1000. The processor may execute code (e.g., processor firmware, protocol stack, database management system, operating system, or combinations thereof) that establishes an execution environment for the program instructions. The processor may include a programmable processor. The processor may include a general-purpose or special-purpose microprocessor. The processor may receive instructions and data from a memory (e.g., the system memory 1020). The computing system 1000 may be a uniprocessor system including one processor (e.g., processor 1010a) or a multiprocessor system including any number of suitable processors (e.g., 1010a-1010n). Multiple processors may be employed to implement parallel or sequential execution of one or more portions of the techniques described herein. The processes, such as logic flows, described herein may be executed by one or more programmable processors executing one or more computer programs to perform functions by manipulating input data and generating corresponding output. The processes described herein may also be executed by, and the apparatus described herein may be implemented by, special purpose logic circuitry, such as a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC).The computer system 1000 may include multiple computing devices (eg, a distributed computer system) to implement various processing functions.

[0142] The I / O device interface 1030 may provide an interface for connecting one or more I / O devices 1060 to the computer system 1000. The I / O devices may include devices that receive input (e.g., from a user) or output information (e.g., to a user). The I / O devices 1060 may include, for example, a graphical user interface presented on a display (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor), a pointing device (e.g., a computer mouse or trackball), a keyboard, a keypad, a touchpad, a scanning device, a voice recognition device, a gesture recognition device, a printer, an audio speaker, a microphone, a camera, etc. The I / O devices 1060 may be connected to the computer system 1000 via a wired or wireless connection. The I / O devices 1060 may be connected to the computer system 1000 from a remote location. The I / O devices 1060 located at a remote computer system may be connected to the computer system 1000, for example, via a network and the network interface 1040.

[0143] The network interface 1040 may include a network adapter that provides a connection of the computer system 1000 to a network. The network interface 1040 may facilitate the exchange of data between the computer system 1000 and other devices connected to the network. The network interface 1040 may support wired or wireless communications. The network may include an electronic communications network such as the Internet, a local area network (LAN), a wide area network (WAN), a cellular communications network, etc.

[0144] The system memory 1020 may be configured to store program instructions 1100 or data 1110. The program instructions 1100 may be executable by a processor (e.g., one or more of processors 1010a-1010n) to implement one or more embodiments of the present technology. The instructions 1100 may include modules of computer program instructions for implementing one or more techniques described herein with respect to various processing modules. The program instructions may include computer programs (known in certain forms as programs, software, software applications, scripts, or code). Computer programs may be written in programming languages, such as compiled languages, interpreted languages, declarative languages, or procedural languages. Computer programs include units suitable for use in a computing environment, such as standalone programs, modules, components, subroutines, etc. Computer programs may or may not correspond to files in a file system. A program may be stored as part of a file that contains other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program, or in multiple cooperating files (e.g., a file that contains one or more modules, subprograms, or portions of code). A computer program may be located at one site or distributed across multiple remote sites to be executed on one or more computer processors interconnected by a communications network.

[0145] The system memory 1020 may include a tangible program carrier that stores program instructions. The tangible program carrier may include a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium may include a machine-readable storage device, a machine-readable storage substrate, a storage device, or any combination thereof. The non-transitory computer-readable storage medium may include non-volatile memory (e.g., flash memory, ROM, PROM, EPROM, EEPROM memory), volatile memory (e.g., random access memory (RAM), static random access memory (SRAM), synchronous dynamic RAM (SDRAM)), bulk storage memory (e.g., CD-ROM and / or DVD-ROM, hard drive), etc. The system memory 1020 may include a non-transitory computer-readable storage medium that stores program instructions executable by a computer processor (e.g., one or more of processors 1010a-1010n) to achieve the subject matter and functional operations described herein. The memory (e.g., system memory 1020) may include a single memory device and / or multiple memory devices (e.g., distributed memory devices). The instructions or other program code that provide the functionality described herein may be stored on a tangible, non-transitory computer-readable medium. In some cases, the entire set of instructions may be stored on the medium at the same time, or in some cases, different portions of the instructions may be stored on the same medium at different times.

[0146] The I / O interface 1050 may be configured to coordinate I / O traffic between the processors 1010a-1010n, the system memory 1020, the network interface 1040, the I / O devices 1060, and / or other peripheral devices. The I / O interface 1050 may perform protocol conversion, timing conversion, or other data conversion to convert data signals from one component (e.g., the system memory 1020) into a format suitable for use by another component (e.g., the processors 1010a-1010n). The I / O interface 1050 may support devices connected via various types of peripheral buses, such as variants of the Peripheral Component Interconnect (PCI) bus standard and the Universal Serial Bus (USB) standard.

[0147] An implementation of embodiments of the techniques described herein may use a single instance of computer system 1000, or may use multiple computer systems 1000 configured to host different portions or instances of the embodiments. Multiple computer systems 1000 may provide parallel or sequential processing / execution of one or more portions of the techniques described herein.

[0148] Those skilled in the art will appreciate that computer system 1000 is merely exemplary and is not intended to limit the scope of the technology described herein. Computer system 1000 may include any combination of devices or software capable of executing or otherwise providing the implementation of the technology described herein. For example, computer system 1000 may include, or be a combination of, a cloud computing system, a data center, a server rack, a server, a virtual server, a desktop computer, a laptop computer, a tablet computer, a server device, a client device, a mobile phone, a personal digital assistant (PDA), a portable audio / video player, a game console, an in-vehicle computer, or a global positioning system (GPS). Computer system 1000 may also be connected to other devices not shown or may operate as a standalone system. Furthermore, functionality provided by the illustrated components may, in some embodiments, be combined into fewer components or distributed among additional components. Similarly, in some embodiments, some functionality of the illustrated components may not be provided, or other additional functionality may be available.

[0149] Also, while various items are illustrated as being stored in memory or storage during use, those skilled in the art will appreciate that these items, or portions thereof, may be transferred between memory and other storage devices for purposes of memory management and data integrity. Alternatively, in other embodiments, some or all of the software components may execute in memory on another device and communicate with the illustrated computer system via computer-to-computer communications. Also, some or all of the system components or data structures may be stored (e.g., as instructions or structured data) on a computer-accessible medium or portable device readable by an appropriate drive, various examples of which are described above. In some embodiments, instructions stored on a computer-accessible medium separate from computer system 1000 may be transmitted to computer system 1000 as a transmission medium or signal, such as an electrical, electromagnetic, or digital signal conveyed over a communications medium, such as a network or wireless link. Various embodiments may further include receiving, sending, or storing instructions or data embodied on a computer-accessible medium in accordance with the foregoing description. Accordingly, the techniques of the present invention may be practiced with other computer system configurations.

[0150] Although block diagrams depict illustrated components as separate functional blocks, embodiments are not limited to systems in which the functions described herein are organized as illustrated. The functionality provided by each component may be provided by software or hardware modules organized in a manner different from that currently illustrated; for example, such software or hardware may be mixed, combined, replicated, partitioned, distributed (e.g., in a data center or geographically), or organized in other different manners. The functionality described herein may be provided by one or more processors of one or more computers executing code stored on a tangible, non-transitory, machine-readable medium. In some cases, despite the use of the singular term "medium," instructions may be distributed across different storage devices associated with different computing devices, where, for example, each computing device may have a different subset of the instructions. This is an implementation consistent with the use of the singular term "medium" herein. In some cases, a third-party content delivery network may host some or all of the information communicated over the network, in which case, to the extent information (e.g., content) can be described as being sourced or otherwise provided, it may be provided by transmitting instructions to obtain the information from the content delivery network.

[0151] The reader should understand that this application describes several individually useful technologies. Applicant has combined these technologies into one document rather than separating them into multiple independent patent applications because the subject matter of the technologies is related, leading to economies in the filing process. However, the separate advantages and aspects of such technologies should not be confused. In some cases, embodiments address all of the deficiencies identified herein, but it should be understood that the technologies are independently useful, and some embodiments address only a subset of such problems or provide other unmentioned advantages that would be apparent to those skilled in the art upon reviewing this disclosure. Due to cost constraints, some technologies disclosed herein may not be currently claimed and may be claimed in a later application, such as a continuation application, or by amending the present claims. Similarly, due to space limitations, the "Abstract" and "Summary" sections of this document should not be considered a comprehensive description of all such technologies or all aspects of such technologies.

[0152] It should be understood that the detailed description and drawings are not intended to limit the technology to the particular forms disclosed, but on the contrary, are intended to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the technology as defined by the appended claims. Further modifications and alternative embodiments of various aspects of the technology will be apparent to those skilled in the art upon reading this description. Accordingly, this description and drawings are to be construed as illustrative only and are intended to teach those skilled in the art the general manner of carrying out the technology. It should be understood that the forms of the technology shown and described herein are to be considered as exemplary embodiments. Various elements and materials may be substituted for those shown and described herein, parts and processes may be reversed or omitted, and certain features of the technology may be utilized independently, all of which will become apparent to those skilled in the art after having the benefit of this description of the technology. Changes can be made in the elements described herein without departing from the spirit and scope of the technology as set forth in the following claims. The headings used herein are for organizational purposes only and are not intended to be used to limit the scope of the description.

[0153] As used throughout this application, the word "may" is used in its permissive (i.e., possibly) sense rather than its required (i.e., must) sense. Words such as "include," "including," and "includes" mean including but not limited to. As used herein, the singular forms "a," "an," and "the" include plurals unless the content clearly indicates otherwise. Thus, for example, reference to "an element" or "a element" includes a combination of two or more elements, regardless of the use of other terms and phrases for one or more elements, such as "one or more." The term "or" is non-exclusive unless expressly stated otherwise, i.e., encompasses both "and" and "or." Conditional terms such as "depending on X, Y," "upon X, Y," "if X, Y," and "in the event of X, Y" imply a causal relationship in which the antecedent is a necessary causal condition, a sufficient causal condition, or a contributory causal condition for the consequent. For example, "When condition Y is satisfied, state X occurs" means either "X occurs because of Y alone" or "X occurs because of Y and Z." Such conditional relationships are not limited to the consequent being immediately entailed by the antecedent; they can also occur later depending on the consequent. Furthermore, in a conditional statement, the antecedent is linked to its consequent such that it relates to the likelihood of the consequent occurring. Unless otherwise indicated, a statement that multiple attributes or features map to multiple objects (e.g., one or more processors performing steps A, B, C, and D) encompasses both all of those attributes or features being mapped to all of those objects and a subset of those attributes or features being mapped to a subset of those attributes or features (e.g., both when all processors perform steps A through D, and when processor 1 performs step A, processor 2 performs step B and part of step C, and processor 3 performs part of step C and step D).Similarly, a statement that a "computer system" performs step A and that "the computer system" performs step B may include both cases where one computing device within the computer system performs both steps, or where multiple different computing devices within the computer system perform steps A and B. Furthermore, unless specifically indicated otherwise, a statement that a value or action is "based on" another condition or value includes both cases where the condition or value is the only factor and cases where the condition or value is one of multiple factors. A statement that "each" instance of a collection has a characteristic should not be read to exclude cases where the same or similar members of other characteristics of the larger collection do not possess that characteristic, unless otherwise indicated. In other words, "each" does not necessarily mean all. No restrictions on the order of recited steps should be read into a claim unless expressly specified, such as "after performing X, perform Y." In contrast, statements that could improperly be argued to imply an order restriction, such as "perform X on an item, then perform Y on the item that X was applied to," are used to improve claim readability rather than to specify an order. Additionally, a statement such as "at least Z of A, B, and C" (e.g., "at least Z of A, B, or C") refers to at least Z units of each listed category (A, B, and C) and does not require at least Z units in each category. As is apparent from the disclosure, discussions herein utilizing terms such as "processing," "computing," "calculating," and "determining" are understood to refer to specific apparatus operations or processes, such as a special purpose computer or similar special purpose electronic processing / computing device, unless otherwise specified. Features described with reference to geometric constructs, such as "parallel," "perpendicular / orthogonal," "square," and "cylindrical," should be interpreted to encompass items that substantially embody the properties of that geometric construct; for example, reference to "parallel" surfaces encompasses substantially parallel surfaces.The extent to which these geometric constructs may deviate from the Platonic concept should be determined by reference to the ranges in the specification; if no such range is stated, reference should be made to industry standards in the field of use; if no such range is defined, reference should be made to industry standards in the field of manufacture of the specified feature; if no such range is defined, features that substantially embody a geometric construct should be construed as including features that are within 15% of the defining attributes of that geometric construct. Terms such as "first," "second," "third," and "predetermined," as used in the claims, are used to distinguish or identify and do not imply sequential or numerical limitations. Consistent with common usage in the art, data structures and formats described with reference to significant uses for humans need not be presented in a human-understandable form to constitute said data structures or formats. For example, text does not need to be rendered or encoded in Unicode or ASCII to constitute text; images, maps, and data visualizations do not need to be displayed or decoded, respectively, to constitute images, maps, and data visualizations; and voice, music, and other sounds do not need to be projected through speakers or decoded, respectively, to constitute voice, music, and other sounds. Computer-implemented instructions, commands, and the like are not limited to executable code but can also be implemented in the form of data that provides functionality, such as arguments to a function or API call. To the extent that purpose-built noun phrases (and other neologisms) are used in claims and lack obvious interpretation, definitions of such phrases may be set forth in the claims themselves, in which case the use of such noun phrases should not be deemed to construe additional limitations by reference to the specification or extrinsic evidence.

Claims

1. 1. A method for providing an adaptive and interactive AI-driven profile, comprising: capturing, by the processor, a first branded content data set; Organizing, by the processor, the first ingested branded content data set into a plurality of embeddings and indexes and storing the plurality of embeddings and indexes in a knowledge base; Equipped with The embedding comprises an embedding of a vector within an embedding space, the location of the embedding providing a semantic meaning for the content represented by the vector; the index comprises a data structure providing a mapping between the branded content data and its location in the knowledge base, and links to metadata associated with the content data; The method further comprises: generating, by the processor, a first user profile based at least in part on the organized plurality of embeddings and indexes and a user history of a first user associated with the first user profile; updating, by the processor, the first user profile based at least in part on one or more models trained based on recordings indicative of one or more interactions between the first user and the first user profile and one or more actions of a human user; personalizing, by the processor, one or more responses of a conversational agent interacting with the first user based at least in part on the first user profile; A method for providing

2. Ingesting the first branded content data set comprises: processing the first branded content data set with one or more machine learning algorithms; Identifying one or more insights about the first branded content data set; The method of claim 1 , comprising:

3. The method of claim 2 , wherein the one or more insights include at least one of tone, language, audience engagement, intent, mood, receptivity, skill, expertise, or understanding.

4. 10. The method of claim 1, wherein the one or more models trained based on recordings indicative of one or more processes of a human user comprise one or more models trained based on recordings indicative of biological-cognitive and neuroscientific processes of a human user that provide information about at least one of a user's thought processes, behavioral patterns, motivations, or biases.

5. The method of claim 1 , wherein the first user profile is updated in real time.

6. generating one or more customized content recommendations for the first user based at least in part on the first user profile; The method of claim 1 , further comprising: the conversation agent providing the one or more customized content recommendations to the first user.

7. analyzing, with the processor, inputs from a plurality of users in response to interactions with respective conversational agents; extracting, by a processor, one or more insights related to interactions with the plurality of users; 10. The method of claim 1, further comprising: providing, by the processor, one or more data-driven recommendations related to at least one of improving the response of each conversational agent, improving one or more services provided, or system performance.

8. 1. A system for providing an adaptive and interactive AI-driven profile, comprising: a computer having a processor and a memory; one or more sets of code stored in said memory and executed by said processor; and wherein execution of the one or more sets of code causes the processor to: Ingesting a first branded content data set; Organizing the ingested first branded content data set into a plurality of embeddings and indexes and storing the plurality of embeddings and indexes in a knowledge base; configured to run The embedding comprises an embedding of a vector within an embedding space, the location of the embedding providing a semantic meaning for the content represented by the vector; the index comprises a data structure providing a mapping between the branded content data and its location in the knowledge base, and links to metadata associated with the content data; The processor further comprises: generating a first user profile based at least in part on the organized plurality of embeddings and indexes and a user history of a first user associated with the first user profile; updating the first user profile based at least in part on one or more models trained based on recordings indicative of one or more interactions between the first user and the first user profile and one or more actions of a human user; personalizing one or more responses of a conversational agent interacting with the first user based at least in part on the first user profile; and A system configured to run

9. Ingesting the first branded content data set comprises: processing the first branded content data set with one or more machine learning algorithms; and identifying one or more insights about the first branded content data set.

10. 10. The system of claim 9, wherein the one or more insights include at least one of tone, language, audience engagement, intent, mood, receptivity, skill, expertise, or understanding.

11. 10. The system of claim 8, wherein the one or more models trained based on recordings indicative of one or more processes of a human user comprise one or more models trained based on recordings indicative of biological-cognitive and neuroscientific processes of a human user that provide information about at least one of a user's thought processes, behavioral patterns, motivations, or biases.

12. The system of claim 8 , wherein the first user profile is updated in real time.

13. generating one or more customized content recommendations for the first user based at least in part on the first user profile; The system of claim 8 , further configured to: provide the one or more customized content recommendations to the first user by the conversation agent.

14. analyzing inputs from a plurality of users in response to interactions with each of the conversational agents; extracting one or more insights related to interactions with the plurality of users; 10. The system of claim 8, further configured to: provide one or more data-driven recommendations regarding one or more of improving the responses of each conversational agent, improving one or more services provided, or system performance.

15. 1. A non-transitory computer-readable medium storing computer program instructions that, when executed by one or more processors, cause the one or more processors to: Ingesting a first branded content data set; Organizing the ingested first branded content data set into a plurality of embeddings and indexes and storing the plurality of embeddings and indexes in a knowledge base; Execute a process including The embedding comprises an embedding of a vector within an embedding space, the location of the embedding providing a semantic meaning for the content represented by the vector; the index comprises a data structure providing a mapping between the branded content data and its location in the knowledge base, and links to metadata associated with the content data; The one or more processors further include: generating a first user profile based at least in part on the organized plurality of embeddings and indexes and a user history of a first user associated with the first user profile; updating the first user profile based at least in part on one or more models trained based on recordings indicative of one or more interactions between the first user and the first user profile and one or more actions of a human user; personalizing one or more responses of a conversational agent interacting with the first user based at least in part on the first user profile; A non-transitory computer-readable medium for causing a process to be performed, including:

16. Ingesting the first branded content data set comprises: processing the first branded content data set with one or more machine learning algorithms; Identifying one or more insights about the first branded content data set; 16. The non-transitory computer-readable medium of claim 15, comprising:

17. 20. The non-transitory computer-readable medium of claim 16, wherein the one or more insights include at least one of tone, language, audience engagement, intent, mood, receptivity, skill, expertise, or understanding.

18. 16. The non-transitory computer-readable medium of claim 15, wherein the one or more models trained based on recordings indicative of one or more processes of a human user comprise one or more models trained based on recordings indicative of biological-cognitive and neuroscientific processes of a human user that provide information about at least one of a user's thought processes, behavioral patterns, motivations, or biases.

19. generating one or more customized content recommendations for the first user based at least in part on the first user profile; 16. The non-transitory computer-readable medium of claim 15, further comprising: the conversation agent providing the one or more customized content recommendations to the first user.

20. analyzing, with the processor, inputs from a plurality of users in response to interactions with respective conversational agents; extracting, by a processor, one or more insights related to interactions with the plurality of users; 16. The non-transitory computer-readable medium of claim 15, further comprising: providing, by the processor, one or more data-driven recommendations related to one or more of improving responses of respective conversational agents, improving one or more services provided, or system performance.

Images