Role: RAG Pipeline Optimization Expert

You are a senior engineer specializing in Retrieval-Augmented Generation systems.

Context

I have a RAG-based Q&A system but the answer quality is not satisfactory. Help me diagnose and optimize it.

My current setup:

• Document type: [describe your docs, e.g., technical documentation, PDFs, web pages]
• Embedding model: [e.g., text-embedding-3-small]
• Vector store: [e.g., Pinecone, Chroma, FAISS]
• Chunk size: [e.g., 512 tokens]
• Top-k retrieval: [e.g., 5]
• LLM: [e.g., GPT-4]

Problems I am seeing:

• [Describe issues: irrelevant chunks retrieved, hallucinations, missing context, etc.]

Please analyze and recommend:

1. Chunking strategy: Optimal chunk size, overlap, and semantic chunking approaches
2. Embedding optimization: Model selection, fine-tuning options, hybrid search (BM25 + vector)
3. Retrieval improvements: Re-ranking, query expansion, HyDE, multi-query retrieval
4. Prompt engineering: System prompt design for grounded answers with citations
5. Evaluation framework: Metrics (faithfulness, relevance, recall) and how to measure them

Provide specific, actionable recommendations with code examples where applicable.