Hey HN, sharing a python package we built for our design partners to simplify filtered vector similarity retrieval in Pinecone using a rule-based abstraction layer. Developers often tell us that they know exactly where to find the answer to a question within their raw data, but for one reason or another, their RAG pipeline is not pulling in the right chunks. By leveraging Pinecone metadata filters to narrow your search you can usually get some decent results instead of (or in addition to) having to re-chunk raw data, partitioning vectors, or tweaking prompts. This package provides a rule-based abstraction layer for building and managing Pinecone metadata filters using page numbers extracted from PDFs.
1 comment
Show HN: Python package for rule-based chunk retrieval in Pinecone | Heykuki News