I build production systems with agentic tools.

I started in Astronomy and then moved to physics. A B.Sc from Universidad Católica taught me to model messy systems, basic programming, electronics and audio technology, and also to trust fundamentals over hand-waving. Somewhere along the way I found I cared more about building than running the experiments.

So I went full-stack. At Pinflag I spent three years on the load-bearing parts of e-commerce: payment integrations, reconciliation, order fulfillment, analytics backends. I learned what production at scale really costs.

At TeselaGen I lead a legacy-API migration to a modern NestJS clean-architecture backend, build the REST and GraphQL APIs and data layers behind an AI copilot for biotech tools, and wrote a Python client SDK that turns the backend's OpenAPI docs into a programmatic client. Around that work I started building an agentic-development toolchain, custom Claude Code skills, MCP servers, and ticket-to-PR workflows.

I ship production code daily with AI coding agents, using spec-driven and test-driven development. I also pair with teammates on system design and on the urgent, unglamorous issues that keep a product healthy. Every feature starts as a written PRD, is implemented against it, and ships only when CI is green and a peer has approved it.

After shipping, I fold the lessons back into the toolchain: the repo context, the skills, and the agent's memory, so the next session starts better prepared. This portfolio is exhibit number one, built in public with that exact loop. Right now I'm experimenting with autonomous loop workflows to push throughput and quality further.

Outside the editor I'm a physics graduate and a former working musician, so I have a soft spot for where math, code, sound, and craft overlap. I build isometric city-builders for fun and use them as testbeds for agentic workflows. I also make sure to go to the gym for some strength training so I can keep lifting my two kids.

I'm based in Santiago, Chile, work in English and Spanish, and care a lot about leaving systems more honest than I found them.