From RAG to fabric: Lessons learned from building real-world RAGs at GenAIIC – Part 1

This blog post discusses techniques to optimize a retrieval-augmented generation (RAG) system for text-based knowledge bases. It covers evaluating RAG performance, troubleshooting issues, and practical tips to improve the retriever component.

Specifically, the article covers:

• Overview of RAG use cases
• Evaluating a RAG solution
• Troubleshooting RAG issues
• A practical guide to improving the retriever:
  • Hybrid search (combining vector and keyword search)
  • Adding metadata to text chunks
  • Small-to-large chunk retrieval
  • Section-based chunking
  • Rewriting the user query
  • Metadata filtering
  • Training custom embeddings
• Improving reliability of generated responses:
  • Prompt engineering guardrails
  • Asking the LM to output quotes
  • Generating answers with inline quotations
  • Verifying quotes
• Conclusion and preview of Part 2 covering RAG with structured data and multimodal inputs