Virtual assistant for generating product based prompt
Inventors
Singh, Ankit • HARTYE, Eric • Subber, Waad • Halpern, Jason
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12657220-B2
Publication Date
2026-06-16
Expiration Date
Abstract
The present disclosure provides a system for providing virtual assistant for generating customized query for retrieving data from a database. The system comprises a processor and a memory storing program instructions, which, when executed by the processor, causes the processor to provide one or more index of categories of the data, for efficient retrieval of the data from the database based on a customer's query. The processor is also configured to provide one or more few-shot prompts for use by the customer for generating an output in a predetermined format corresponding to the customer's query. The processor is configured to receive a query from the one or more customer, wherein said query is a text and is based on at least one of index of categories and the few-shot prompt. The processor, based on indexes and few-shot prompt processes the query using natural language processing to extract one or more textual content from the query and sends the query to the database for retrieval of the data. The system retrieves the relevant data based on identified textual content, said relevant data is based on semantic similarity between the vector index and displays the retrieved data to the one or more customer based on the predetermined format specified by few-shot prompt.
Core Innovation
The invention provides a virtual assistant for generating a customized query for retrieving data from a database. The system creates one or more index of categories of data that are specific to one or more user and provides the one or more category indexes for efficient retrieval of the data from the database based on a user's query. It also provides one or more few-shot prompts for use by the user for generating an output in a predetermined format corresponding to the query.
The virtual assistant receives a query from the one or more user, where the query is a text based on at least one of the index of categories and the few-shot prompt. The system processes the query using natural language processing to extract one or more textual content from the query. It then sends the query to the database for retrieval of the data and retrieves relevant data based on identified textual content, where the relevant data is based on semantic similarity between the vector index.
The system displays the retrieved data to the one or more user based on the predetermined format specified by the few-shot prompt. The document further describes implementations involving hierarchical indexing or flat indexing for the index of categories of data, and includes use of vector embeddings stored in the database. It also describes updating the categories index when a vector index of data changes, and training an AI model using a selected index of categories from the data and a few-shot prompt provided by one or more users.
Claims Coverage
The document contains three independent claims: clm-00001 (system), clm-00009 (method), and clm-00015 (non-transitory computer-readable storage medium). Across these independent claims, the coverage centers on a virtual assistant that uses user-specific category indexes and few-shot prompts to generate a customized text query, processes the query with natural language processing, retrieves semantically similar data from a database via a vector index, and displays the retrieved data in a predetermined format.
User-specific category indexes for efficient retrieval
create one or more index of categories of data specific to one or more user; provide the one or more index of categories of the data, for efficient retrieval of the data from the database based on a user's query; and receive a query from the one or more user, wherein the query is a text and is based on at least one of the index of categories and the few-shot prompt.
Few-shot prompts for predetermined output formatting
provide one or more few-shot prompts for use by the user for generating an output in a predetermined format corresponding to the query; and display the retrieved data to the one or more user based on the predetermined format specified by few-shot prompt.
Natural language processing textual content extraction and semantic similarity retrieval via vector index
process the query using natural language processing to extract one or more textual content from the query; send the query to the database for retrieval of the data; and retrieve the relevant data based on identified textual content, where the relevant data is based on semantic similarity between the vector index.
User-specific virtual assistant implemented as program instructions
storing program instructions providing a virtual assistant for generating a customized query for retrieving data from a database, the instructions, when executed, perform the steps of: creating one or more index of categories of data specific to one or more user; providing the one or more index of categories of the data, for efficient retrieval of the data from the database based on a user's query; providing one or more few-shot prompts for use by the user for generating an output in a predetermined format corresponding to the user's query; receiving a query from the one or more user, wherein the query is a text and is based on at least one of index of categories and the few-shot prompt; processing the query using natural language processing to extract one or more textual content from the query; sending the query to the database for retrieval of the data; retrieving the relevant data based on identified textual content, where the relevant data is based on semantic similarity between the vector index; and displaying the retrieved data to the one or more user based on the predetermined format specified by few-shot prompt.
Across the independent claims, the core inventive coverage combines creating and using one or more index of categories of data specific to one or more user, providing few-shot prompts that drive a predetermined output format, processing a user-provided text query using natural language processing to extract textual content, and retrieving semantically similar relevant data from the database using a vector index, followed by displaying the retrieved data according to the predetermined format.
Stated Advantages
Efficient retrieval of the data from the database based on a user's query.
Documented Applications
Example use cases are described for customers generating customized text queries to retrieve product-related information from a (vector) database and generating/delivering product reports.
Interested in licensing this patent?