I did time at FAANGs as a SWE and then Forward Deploy Engineer / Partner Engineer, a recent stint in VC, some stealthy startup time. Now I build apps with my AI interns, with a lot of quality checks. I'm the engineer who groks the business. Ship the thing that moves the number.
Three models compete in a custom evaluation harness. The winner drives inventory recommendations, with monitoring and a reproducible notebook. Built in under seven days with my own Python harness.
Estimates VRAM and recommends GPUs for inference, embeddings, vision, image diffusion, and fine-tuning based on task, model size, quantization and batching.
Verification infrastructure for AI coding agents. LoopGate runs agent-written code through repo-specific checks before accepting output.
Gates for the browser, not just the build. A TypeScript frontend harness for coding agents. This site itself was built using loopgate_js.
For teens, for fun. A tiny weather-to-outfit app for what to wear today. Enter a ZIP code, check weather, get a weather-specific outfit.
Three AIs walk into a frontier lab. This is what happened.
A quick attempt to answer the forever debate, rent or own?
Thoughtful adoption, not hype. Minimalism, not line count. Evidence, not guesses. And moves regardless of the crowd. Curious about: AHI (agent-human-interaction), the evolution of evaluations, and the practical mess of using AI to build.
repos and writeups
