Blog

Why I ripped free-form LLM codegen out of my migration pipeline

Building a multi-agent PHP-to-Laravel migrator, I rewrote the generation layer twice. The second pass stopped asking the LLM to write structural code — and got faster, cheaper, and more reliable for it.

24 Apr 2026 · 4 min read

Why I chose Groq for PreTriage — and what I'd use in production

Speed isn't just a performance metric in a clinical context — it's a UX requirement. Here's how I picked an inference provider for a real-time AI triage app, and what I'd do differently at scale.

7 Apr 2026 · 4 min read