Abstract

Methods related to retrieval-augmented generation are disclosed.  A computer system can receive a requestor query.  The computer system can identify one or more context data elements (comprising e.g., “chunks” of documents) that are relevant to the requestor query based on header-weighted embeddings.  The computer system can generate a prompt by combining the one or more context data elements and the requestor query.  The computer system can generate a requestor query response based on the prompt and cause the requestor query response to be provided to the requestor.  The computer system can additionally collect various performance metrics, which can be used to identify candidate context data elements to be “rechunked” as part of a rechunking process.  The computer system can determine rechunking policies corresponding to the candidate context data elements and rechunk data records based on those chunking policies, thereby generating new context data elements, e.g., using a large language model.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS