Notes from production. We write about the things we'd want to read — eval frameworks, RAG that survives a Series B, retailer ingestion, design systems across platforms, and the boring parts that determine whether a system actually works.
How Lexly ships legally binding contracts across six legal frameworks from a single LLM pipeline — without asking the model to invent the law. Retrieval, structured outputs, and the eval harness that keeps it honest.
Read the postMost AI systems we replace have one eval — a notebook someone ran once. Here is the structure we use on every project to make the eval an artefact that survives upgrades to the underlying model, prompts, retrieval and product surface.
Read post →Most deal aggregators are held together with cron jobs and goodwill. The retailer side is the real product. Here is the structured ingestion pipeline that lets TopPrix retailers update their own listings on a Sunday night without ever opening Slack.
Read post →How we shipped iOS, Android and a web app off one design tokens package, one product team and one Friday demo cadence — and why that decision saved the project six months.
Read post →