AI1 min read

Retrieval is a data problem before it is a model problem

Most disappointing RAG systems are not failing at generation. They are failing at what was retrieved, and no amount of prompt tuning repairs that.

By Optirix Admin

When an assistant answers badly, the instinct is to change the prompt or the model. In our experience the fault is nearly always upstream.

Look at the chunks

Before touching generation, read what was actually retrieved for a failing question. If the answer is not in there, the model was never going to succeed.

Evaluate retrieval separately

Score retrieval on its own — did the right passage make the top five? — so you can tell a chunking problem from a reasoning problem.

  • #ai
  • #rag
Next step

Have somethingambitious in mind?

Let’s turn your next idea into something people remember. Send us the rough version — we are used to shaping those.