Building AI Features That Survive Production: Start With Evals, Not PromptsAI Engineering · 3 min readThe teams I see struggling with AI features usually have the same problem: they can show you a great demo, but they cannot tell you whether last Tuesday's prompt change made anything better or worse.RAG Is a Data Engineering Problem Wearing a Trench CoatAI Engineering · 3 min readMost broken RAG systems I have debugged were not model failures. They were document pipelines that fed the model stale, duplicated, or badly sliced text and then blamed the answers.AI Coding Assistants Made Review the BottleneckAI Engineering · 3 min readAssistants made first drafts nearly free, and pull requests quietly doubled in size. The constraint in software delivery was never typing speed, and now everyone can see it.Observability for LLM Features Without Drowning in LogsAI Engineering · 3 min readThe useful question is never what the model said word for word. It is which version answered, how long it took, what it cost, and whether the user got what they came for.