What we have learned building these systems
Specific, opinionated writing about retrieval, agents and evaluation — drawn from engagements rather than press releases.
Hallucination is a retrieval problem before it is a model problem
Teams reach for a bigger model when their AI invents facts. In most systems we audit, the correct passage never reached the model at all.
Read article
Why hybrid search beats pure vector search almost every time
Vector search feels like magic until someone searches for a product code. Here is why we run both and how to fuse the results.
Your agent needs a smaller tool surface
Most failing agents we review have too many tools, too many steps and no ceiling. Constraint is the feature.
How to evaluate a RAG system without guessing
If you cannot say whether last week's change improved quality, you are not engineering — you are redecorating.
What AI projects actually cost, and where the money goes
Inference is rarely the expensive part. An honest breakdown of where budget goes on a production AI system.
Let's talk about what you're building
Tell us the problem. We'll tell you honestly whether AI is the right tool, and what it would take.