Agentic RAG for Code & Engineering Knowledge

You are an AI applications engineer. Build an agentic retrieval-augmented generation system for code and engineering knowledge. Tech Stack: [INSERT - Python + LangGraph/LlamaIndex + FastAPI] LLM: [INSERT - GPT-4o / Claude 3.5 / local Qwen2.5-Coder] Sources: [INSERT - Git repositories, docs sites, wikis, ticket trackers, runbooks] Vector/Graph Store: [INSERT - Milvus/Qdrant, optional Neo4j] Deliver the following modules in order: 1. Architecture: a planner agent decomposes questions into retrieval steps; tool-using research agents iterate search-read-synthesize until the question is answered. 2. Indexing Data: code chunks with symbol metadata, docs with heading hierarchy, tickets with status; hybrid vector + BM25 + graph (symbol/call) indexes; incremental sync. 3. Retrieval APIs: multi-hop query, query rewriting/HyDE, re-ranking, time and repository filters, citation handles returned per chunk. 4. Agent Loop: tool calls for search/read_file/grep, self-reflection on evidence gaps, answer synthesis with file/line citations and abstention when unsupported. 5. Feedback: thumbs up/down, unresolved-question queue, automatic re-indexing triggered by feedback signals. 6. Testing & Evals: labeled multi-hop QA set, citation correctness and faithfulness metrics, recall@k benchmark, adversarial outdated-doc tests.