A few weeks ago I got rejected from a role and the feedback was that I needed a deeper understanding of FAISS and LlamaIndex.

So I built triage.flow — an AI assistant that lets you explore and understand GitHub repositories through a chat interface.

It clones a repo, indexes it using FAISS + BM25 + tree-sitter parsing, and powers a full UI where you can:

    - Ask natural-language questions like “how does auth work?” or “explain @src/components/Modal.tsx”

    - Mention specific files/folders with @filename.ts (autocomplete supported)

   - See how the agent thinks in real time (Thought → Action → Observation)

   - View retrieved code side-by-side with the chat

It uses a hybrid RAG system, ReAct-style agent reasoning, and streaming responses — built with FastAPI, LlamaIndex, and React.

GitHub: https://github.com/ashikshafi08/triage.flow

Would love feedback or thoughts from anyone building similar tools.

(Also exploring new opportunities — open to roles around LLM infra, RAG, or applied ML.)