Essays on the engineering decisions that decide whether a RAG system is fit for regulated work. Long-form, opinionated, with worked examples from real builds.
The single largest jump in retrieval accuracy on the Sarawak Labour Law system came from re-chunking on subsection boundaries. The embedding model never changed. Here is what I tried first, why it failed, what I changed, and what I would still do differently.
Most RAG projects skip evaluation. This is how I do it on legal builds — the three axes that matter, how to construct a test set without going crazy, and how to use LLM-as-judge without fooling yourself. Includes a link to the open-source harness.
azamzam.ai