The RAG system is returning irrelevant answers even though relevant documents exist in the vector database.

[shivamxverma]

1. Send query: "What is refund policy?"
2. Embedding generated using OpenAI
3. Top-3 documents retrieved from vector DB
4. Context injected into prompt
5. LLM response generated