Section
AI / ML
Stack
RAG · LangChain · ChromaDB · OpenAI
Category
RAG · LLM Application
Status
Pipeline complete, evaluation and chunking tuning ongoing
01 — Challenge
The core engineering challenge
LLMs hallucinate confidently on domain-specific questions — grounding generation in retrieved evidence with explicit citations shifts the failure mode from silent hallucination to visible retrieval gaps.
03 — Decisions
Engineering decisions and what they cost
01Chunk documents by semantic boundary rather than fixed token count to preserve context across splits.
02Return source passages alongside generated answers so users can verify every claim.
03Separate retrieval quality evaluation from generation quality — they fail in different ways.
04 — Stack
RAGLangChainChromaDBOpenAIEmbeddingsPython