Review My RAG Pipeline

AI engineering ragretrievalarchitecture Updated June 2026

Get a structured review of a retrieval-augmented generation design.

Why it works: The retrieval-vs-generation diagnosis question is the key move; most RAG debugging fails by changing both at once.

This is a message template. Fill in the blanks below, then paste it into the chat.

---

Review my RAG pipeline design. Be direct about weaknesses; I’d rather find them now than in production.

Use case: [what questions users ask and what documents you’re retrieving from]

Current design:

• Chunking: [strategy and chunk size]
• Embedding/retrieval: [model, vector store, top-k, any hybrid/keyword search]
• Reranking: [if any]
• Generation: [model, how retrieved context is presented in the prompt]

Symptoms I’m seeing: [e.g. “answers miss obviously relevant docs”, “model ignores retrieved context”, “works on short queries, fails on multi-part ones”, or “not built yet, reviewing the design”]

Please review:

1. Retrieval quality risks: where relevant content is most likely to be missed given this chunking and search setup
2. Generation risks: where the model is likely to ignore, misuse, or contradict the retrieved context
3. The diagnosis I should run first: how to tell whether my problem is retrieval or generation before changing anything
4. What I’d cut: any complexity in this design that isn’t earning its keep

Fill in the blanks

---