Code Review Assistant with Repo-Aware RAG
A review assistant that grounds LLM suggestions in the actual codebase, past PRs, and architecture decisions.
LangChainOpenAI APIFastAPIFAISS

Highlights
- AST-aware chunking and symbol-graph retrieval for cross-file context.
- Change-impact summaries derived from call-graph diffs.
- Inline PR comments with reviewer overrides feeding an evaluation loop.
Outcomes
- Reduced mean time to first review by 42% on active repos.
- Caught 3.1x more regression risks than a plain diff-only LLM baseline.
Stack
- LangChain
- OpenAI API
- FastAPI
- FAISS
- PostgreSQL