Abstract: Systems and methods for integration of anchor information for retrieval augmented generation are provided. A communication platform accesses communication data associated with an enterprise on the communication platform. The communication platform divides the communication data into multiple data chunks. The communication platform generates one or more extended queries for each data chunk of the multiple data chunks using a generative AI model. The communication platform receives a user query, and identifies, from the multiple data chunks, one or more relevant data chunks for the user query at least by comparing the user query with the one or more extended queries associated with each data chunk of the multiple data chunks. The communication platform provides an answer to the user query based on the one or more relevant data chunks.

Abstract: In some implementations, a data partner may transmit a first request associated with a user account and may receive a first response including an access token. The data partner may transmit a second request, that includes the access token, for a list of applications associated with the user account and may receive a second response including the list of applications. The data partner may transmit a third request, that includes the access token, for a list of activities associated with the list of applications and may receive a third response including the list of activities. The data partner may transmit a fourth request, that includes the access token, to unlink a selected application, from the list of applications, from the user account. The data partner may receive an indication that a token associated with the selected application was revoked.

Abstract: Techniques for implementing hybrid retrieval augmented generation for rich document queries using a large language model (“LLM”) are disclosed. In an example method, a computing system receives documents which are added to databases. The computing system receives a query, from which tokens are generated. The computing system determines a first ranking of the documents using a probabilistic ranking method and a second ranking of the documents in response to a vector database query using an embedded representation. The computing system determines a third ranking of the documents based on the first ranking and the second ranking. The computing system selects top documents from the third ranking. The computing system determines a knowledge graph based on the top documents. The computing system outputs the query, the top documents, and at least a portion of the knowledge graph to an LLM. The computing system receives and outputs a response.

Abstract: Systems and methods of evaluating and fine-tuning a generative AI tool on a communication platform. The communication platform accesses a dataset comprising a user query and a response generated by an AI-based query system. The communication platform evaluates the response with respect to the user query using multiple AI-based scoring models to obtain multiple evaluation results. In response to determining that the multiple evaluation results are inconsistent, the communication platform evaluates the response with respect to the user query using a reference large language model (LLM) to provide a reference evaluation result. In response to determining that the reference evaluation result is decisive, the communication platform classifies, based on the reference evaluation result, the dataset to a data category of one or more data categories. The communication platform fine-tunes the AI-based query system based on a group of datasets in the data category.

Abstract: Systems and methods of knowledge graph construction are provided. A communication platform accesses communication data on the communication platform associated with an enterprise. The communication platform extracts a plurality of keywords from the communication data. The communication platform creates a knowledge graph comprising a plurality of nodes. The plurality of nodes comprises a plurality of keyword nodes. The communication platform identifies, using a graph random walk algorithm, one or more keyword nodes in the knowledge graph related to a first keyword node to link the first keyword node with the one or more keyword nodes. The communication platform determines a first keyword node embedding for the first keyword node by aggregating information associated with at least the first keyword node and the one or more keyword nodes using a graph convolutional network algorithm. The communication platform attaches the first keyword node embedding to the first keyword node in the knowledge graph.

Abstract: Example methods and systems for context-based channel recommendation and management are provided. A communication platform creates a user embedding representation for a user based on a set of user data associated with the user on a communication platform. The communication platform creates a plurality of channel embedding representations for a plurality of chat channels on the communication platform based on channel data associated with the plurality of chat channels. The communication platform determines a similarity score corresponding to each of the plurality of chat channels with respect to the user based on the user embedding representation and the plurality of channel embedding representations to obtain a plurality of similarity scores. The communication platform generates a recommendation of one or more chat channels to the user based on the plurality of similarity scores. The communication platform provides the recommendation of one or more chat channels to the user.

Abstract: Techniques are disclosed for implementing a multi-agent framework for generating responses to queries. In an example method, a computing system receives, from a client device, a query. The computing system determines one or more query portions based on the query. The computing system determines a domain agent for each query portion, in which each domain agent is configured to determine execution steps for one or more executors to cause the executors to generate a response to the query portion. The computing system provides first information about the one or more query portions and the determined execution steps to the respective determined domain agents. The computing system receives, from the one or more domain agents, response portions. The computing system generates the response using the response portions. The computing system outputs, to the client device, the response to the query.

Abstract: Techniques are disclosed relating to a multi-agent search client. In an example method, a search agent application receives a query. The search agent application determines, using a search agent, one or more agents to each execute a portion of the query. The search agent application outputs, to a domain agent, a portion of the query. The search agent application determines, using the domain agent, an execution step based on the portion of the query and information about an executor. The executor generates instructions based on the execution step that are configured to generate a response to the execution step and outputs a command to execute the instructions. The executor then outputs, to the first domain agent, the response to the execution step. The search agent application outputs, using the search agent, a response to the query based on the response to the execution step.

Abstract: Techniques are disclosed relating to a multi-agent search client. In an example method, a search agent application receives a query. The search agent application determines, using a search agent, one or more agents to each execute a portion of the query. The search agent application outputs, to a domain agent, a portion of the query. The search agent application determines, using the domain agent, an execution step based on the portion of the query and information about an executor. The executor generates instructions based on the execution step that are configured to generate a response to the execution step and outputs a command to execute the instructions. The executor then outputs, to the first domain agent, the response to the execution step. The search agent application outputs, using the search agent, a response to the query based on the response to the execution step.

Abstract: In some implementations, a data partner may transmit a first request associated with a user account and may receive a first response including an access token. The data partner may transmit a second request, that includes the access token, for a list of applications associated with the user account and may receive a second response including the list of applications. The data partner may transmit a third request, that includes the access token, for a list of activities associated with the list of applications and may receive a third response including the list of activities. The data partner may transmit a fourth request, that includes the access token, to unlink a selected application, from the list of applications, from the user account. The data partner may receive an indication that a token associated with the selected application was revoked.