Ask HN: RAG-Based PR Review System – Looking for Boilerplates or Alternatives

I’m working on building a RAG (Retrieval-Augmented Generation) system for my codebase to enable a PR review system. Currently, I’m using LangChain with ChromaDB as the vector database. I parse the code using AST and generate embeddings with a HuggingFace sentence transformer.

While it appears to index the code chunks, the query results are often not very relevant.

Is there any boilerplate code available for this use case? Or are there any alternative approaches or tools that might work better?