Category
Retrieval, evaluation, and the rest of the engineering discipline that sits around a model once it has to run in production.
Everyone talks about the model. Almost nobody talks about the pipeline that feeds it, the eval that catches when it drifts, or the failure modes that only show up once real traffic hits it — and that gap is where most production AI work actually happens.
This category is about that surrounding engineering: retrieval, evaluation, and the discipline of building systems around a model rather than just calling one.
Getting the right passage into the context window — chunking, indexing, and the search problem hiding underneath every RAG pipeline.
Most RAG debugging starts at the prompt and the model, when the failure usually happened earlier — at chunking. This post runs six chunking strategies against the same 400-page corpus and scores each one on whether the right passage actually made it into the context window, with the eval harness and the golden set published alongside the results.
Writing the eval before the prompt — treating an AI feature's evaluation suite as the spec rather than an afterthought.
Writing the prompt before the eval is the same mistake as writing code before the test — it feels faster and costs more later. This post makes the case for treating the eval as the spec an AI feature is built against, and shows what teams actually pay when they add it after the feature has already shipped.