A pip-installable CLI that fuses classic keyword search with semantic embeddings to find code the way you actually think about it.

Why I built it: I kept hunting for small pieces of code across large codebases and keyword search alone wasn't cutting it. AI tools could find it, but at a real cost in tokens for something a cheaper, purpose-built tool could do instead. So I built my own: chunking code, indexing it as vector embeddings, and combining semantic search with keyword matching to find what I was actually looking for.

• Built a pip-installable hybrid code search CLI that improves retrieval relevance over standard baselines by fusing BM25 keyword search and Voyage AI embeddings via Reciprocal Rank Fusion.
• Increased embedding precision by designing a tree-sitter AST parsing pipeline that chunks multi-language source code (Python, JavaScript, TypeScript) strictly at the function and class levels.
• Eliminated redundant indexing overhead by implementing a git-aware post-commit hook that incrementally re-embeds only modified files while safely extending existing repository hooks.