Senior full stack engineer · San José, Costa Rica
I've been building for the web for over 10 years: agency work for brands like American Express and PlayStation, then product teams at Gorilla Logic. Now I build software with AI agents and still know exactly what got built, because I verify everything they build.
Gorilla Logic · Prodigious / Publicis · Digitas · Amex · PlayStation · Delta
This site runs on one small cloud server that I set up myself. Here is what happens between your browser and the Node.js process, and the five jobs Nginx does in the middle.
I had to walk my team through a login change that touches several applications. Instead of scrolling through the diff, I had an agent turn the old and new flows into a small interactive machine.
I installed Rust and built dirsize, a small CLI that shows what is eating your disk as a sorted tree. It reads 153,000 files in about a quarter of a second.
Sits between an app and an LLM API: masks names, SSNs, card numbers and other PII with typed placeholders, then restores the real values in the reply and tool calls. Per-category mask, block or allow policy, an audit log that never stores values, and zero runtime dependencies.
Benchmarks TypeSafe's Jev against Claude Haiku 4.5 and Opus 5 on 72 labeled decisions, with the same questions and scoring code for every model. Jev scored 94.4% at ~185 ms and ~$0.02 per 1k calls; the report also covers calibration (ECE, Brier).