How we build, written down.
Reference architectures are design documents, not client projects. When we publish a measurement, we ran it. When we publish a target, we say it is a target.
Internal work.
Labelled as internal, because that is what it is.
How we would build retrieval over a document estate that already has an access model, and how we would know whether it works.
Read it → Agent patterns, and how each one failsThree shapes that cover most of what people mean by an AI agent, each with the failure mode that decides whether it reaches production.
Read it →Still to come.
Measured experiments on our own systems: retrieval quality, chunking, cost per query. Labelled as internal research, because that is what they are.
Client work, published only with the client's agreement. There are none yet, and inventing one would be the fastest way to lose the argument this whole site makes.