Insights
Notes from the field.
Things we have learned building systems that had to keep running. Written by the people who built them, which is why some of it is inconveniently specific.
Serving a 72B open-source model under concurrent load
Batching raises throughput and costs you latency. Choosing between them is most of the job.
AI Engineering · 7 min read
What an AI readiness assessment actually covers
Four weeks, six questions, and one honest answer about whether to build at all.
Consulting · 6 min read
Adding an AI feature to a product that already has users
The constraint is never the model. It is everything the model has to fit into.
AI Engineering · 6 min read
RAG or fine-tuning: how we choose for a mid-market SaaS
The question is usually framed as a technical one. It is almost always an operational one.
AI Engineering · 5 min read
What thirteen years of healthcare software taught us about compliance-first design
Compliance treated as a final review is a rewrite. Treated as a design input, it is mostly free.
Engineering · 6 min read
Got a question one of these did not answer?
We would rather have the conversation than write another article about it.