The scenario
CodeCompass clones a repository, splits code by functions and classes, embeds it, and answers questions with references to the exact files. A great way to learn code-aware chunking and retrieval.
System architecture
01 GitHub repo Clone
02 Code chunker Functions + classes
03 Vector index Chroma
04 Chat Answers with file links
High-level flow of CodeCompass, from input to output.
What you will learn
- Syntax-aware code chunking
- Retrieval over source code
- Linking answers to files and lines
What you will build
- Repo loader
- Code index
- Chat UI with file links
Build plan
- Loader
- Chunker
- Index
- Chat UI
- Deploy