While most RAG demos work in the sandbox, an experimental study from the National Innovation Centre for Data found that only 11% of RAG answers to complex questions are fully correct. Without a repeatable way to test accuracy, your team likely won’t find out if RAG falls apart in production until users complain.

Join this technical session to design a benchmark set that assesses RAG performance on different question types, so you’ll know whether a change improved the system before it ships. You’ll be able to determine exactly where changes to your pipeline yielded stronger results: in retrieval, prompting, models, or another area.

Learn how to:

• Evaluate a RAG pipeline end-to-end and score it, live
• Build a benchmark that evolves with a complex pipeline, adding new questions and edge cases
• Verify facts to achieve better accuracy and production-quality RAG

You’ll walk away with a comprehensive understanding of how to measure RAG systems for accuracy in answering a variety of commonly asked question types.