Technical deep-dives into AI Engineering, LLMs in production, and lessons learned building real systems.
53% to 87% accuracy — without touching the model. This is the story of building an eval-driven LLM pipeline with prompt versioning, a golden dataset, and Braintrust. And why the cheaper model tied the expensive one.
Five months in production. Over 3,000 court notifications processed. 99.22% success rate. This is the story of taking a Legal AI system from chaos to production.
A founder with an Excel spreadsheet and a fuzzy vision. One full-stack engineer. Two weeks. This is the story of how I turned a scattered intake conversation into an executable Sprint 1 — and why what I didn't build matters as much as what I did.