Semantic Search Over a Codebase
Find code by what it does rather than what it is called.
Intermediate10–16 hours estimatedRAG
The problem
Grep finds names. It does not find "the place we handle refund edge cases" when that function is called `processReversal`. Code is a genuinely different retrieval problem from prose, and treating it like prose is why most attempts disappoint.
Who uses it: A developer new to a large repository trying to find where something happens.
Before you start
- Python
- A repository large enough to get lost in
Stack
Pythontree-sitterAn embeddings modelpgvector or Chroma