Contracts before code
The interface between two systems is the thing that outlives both of them. I would rather spend a day on the shape of an API than a week on what happens when two services disagree about it.
About
I turn complicated domains into software people can actually operate
I'm an AI Software Engineer. I work across the whole path from interface to data — product interfaces, BFFs, event-driven microservices — and I build retrieval and agentic systems with the same engineering discipline: contracts, tests, guardrails, observability. If you're looking for someone who can own a problem end to end, that's the part I enjoy most.
Gonzalo Martin Perez — AI Software Engineer, Bahía Blanca, Argentina (UTC-3). 2 years and 8 months, since January 2024.
My route in was a five-year engineering degree at Universidad Nacional del Sur, alongside independent client work that started in 2024 and has not really stopped. The through-line is product engineering: I like owning a thing from the interface down to the queue that makes it work.
Since then I have worked on a promotions platform at fintech scale, a permissions back office spanning an enterprise's internal systems, and an exam platform that five institutions depend on. Applied AI runs through the recent work rather than sitting beside it — a retrieval assistant inside a merchant product, agent pipelines for clients, and the agentic engineering environments I build to do the work itself.
I am currently an AI Engineer at Rampy, and that is the direction I am deliberately going deeper in: agents, retrieval, and the engineering discipline that makes them dependable rather than impressive once.
How I work
Four things I am opinionated about
Correct under concurrency, not just under demo
Most of what I have built has real simultaneous users — exams being sat, promotions being ingested, permissions being checked. Concurrency is a property of the data model, so that is where I put the effort.
Measure before claiming
Every figure on this site carries how it was observed, or says plainly that it is an estimate. A number without conditions is decoration, and I try not to publish decoration.
AI held to the same standard
Retrieval and agents are software. They get the same treatment as everything else: typed boundaries, guardrails, tests on the paths that matter, and observability so a bad answer can be traced rather than guessed at.
What I am looking for
Remote work where engineering and product meet
I'm most useful where the technical problem and the product problem are the same problem, and I can start immediately. If that sounds like your team, the work is one click away — and so is my inbox.