shubhamharkare/codesearch-ai

CodeSearch AI answers questions about React documentation, grounding responses in retrievable sections and declining unsupported questions. The user workflow runs from the Streamlit client through the FastAPI ask route into the answer pipeline: optional follow-up condensation and cache lookup, retrieval/reranking, abstention, then local LLM generation and citation validation. A separate ingestion workflow fetches and chunks the documentation into a versioned dense/BM25 index. Admin endpoints expose health and index status. The README documents these main flows; monitoring and explanation modules are present, but their runtime wiring was not sampled.