About

I'm a software engineer from Mexico. I lead engineering for a ~15-person team building across telematics hardware, GPS services and the software connecting them, and I work end to end: architecture, data, agents and the interfaces people use to operate them. Most of it sits at the seam where hardware, protocols, LLMs and unit economics meet. That seam is where systems usually break.

AI is the part I enjoy most right now. Not the demos — the engineering underneath them: how much context a model actually needs, where reasoning earns its cost, how to keep an agent inside a budget, and what a system should do when the model is confidently wrong. And I spend an unreasonable amount of time benchmarking language models.

How I build

• Architecture before hype. I understand the workload before choosing the technology, instead of forcing every problem into the same architecture.
• Boring but debuggable. I'd rather ship something I can reason about at 2am than something elegant I can't.
• Separate what changes at different speeds. Coupling protocol, product and pricing layers is how you get systems nobody can touch.
• Observability is part of engineering. Production behavior matters more than assumptions.
• AI where reasoning actually helps. Not every problem needs an LLM; this is almost always better than Everything → LLM:

Algorithms → Detection → Structured Context → AI Reasoning

Infrastructure I run

Running my own metal keeps me honest about what managed platforms are actually charging for.

• Proxmox — two nodes, ZFS, UPS via NUT, nightly backups of 38 production databases
• Zero-trust networking — a NetBird mesh instead of firewall holes
• Event infrastructure — Redpanda, Centrifugo, Inngest
• Observability — SigNoz, distributed tracing, P95/P99 latency
• Self-hosted GPS — a production Traccar instance