I build things that matter, with the right people.

Unified Payment Processor.

An in-house Stripe for Kavak, built from scratch.

I designed and built a proxy with its own database and Factory-pattern wrappers that integrated six payment providers (Stripe, Conekta, PayPal, MercadoPago, EBANX, and PayU) behind a single uniform API. The idea was simple on paper but ambitious in practice: turn launching a new country from a months-long project into a two-week integration.

100% of Kavak's money ended up flowing through this system — reservation checkouts, vehicle purchase wires, payments with revenue split to third parties (Stripe Connect-style), subscriptions, reconciliations, reporting. At its core, an in-house Stripe built for commercial reasons and to optimize fees.

Kavak Total.

A product that internally could have been a company.

I led the conception and technical build of Kavak's extended warranty and post-sales benefits product — from architecture all the way to production launch. The squad grew to fifteen people across backend, fullstack, iOS, Android, QA, and DBA, and we worked against commercial deadlines that left little room for mistakes.

Internally it came to be considered a business line that could stand on its own. The CEO and PM publicly recognized it as one of the most successful launches of the year — but what I learned most was how to coordinate an operation of that size without losing technical quality along the way.

Pacto Box · edge computing.

A restaurant POS that keeps running even when the internet drops.

I led the development of an offline-first appliance — a Dell mini-PC running Linux and Docker — meant to run the restaurant POS without depending on internet connectivity. I worked from the ISO image and the boot process up through the internal networking (custom DNS, in-house WiFi AP, traffic bypass) and an atomic edge-to-cloud sync system on top of a serverless event broker.

The architecture allowed cloud branches and edge branches to coexist without conflicts — a small detail that, in real operation, was the difference between a usable system and one that caused more problems than it solved. I showcased the prototype at ABASTUR 2023, Mexico's largest hospitality industry trade show.

Proposal Tool · SaaS platform.

OCR, scraping, simulation, finance, AI — and under one thousand dollars a month in infrastructure.

It's the centerpiece of Solara's solar and BESS financing platform. It combines several data sources to generate commercial proposals in minutes — OCR of CFE bills (Mexico's national power utility) on Textract with a model trained to 98% accuracy, in-house scraping of the CFE portal for consumption history, generation simulation with PVWatts and our own math, financial modeling engine, and final proposal generation as a white-label PDF plus an interactive web experience with an AI chat on Bedrock using Anthropic and Meta models.

Platform transformation · EdTech.

Taking a legacy platform to a domain-driven architecture — with a team built to pull it off.

At Luca — Latin America's #1 EdTech and top 3 worldwide according to Google — I came in as Head of Engineering reporting to the CEO, owning technology, product, and design. The team came first: I restructured the organization, hired senior and junior profiles, and consolidated a group of twenty people. In parallel I introduced a function the company didn't have — the platform team — with specific people taking real ownership over infrastructure and data. That made it possible to bring all infrastructure under Terraform and then optimize it, cutting the GCP bill by 60% while closing gaps in security, stability, and the absence of monitoring.

On that foundation we took a legacy project — coupled, poorly normalized, hard to scale and maintain, and incompatible with the long-term roadmap — to a domain-driven architecture with versioning across every layer. I organized the team into autonomous cells working under a high-rotation Kanban on Linear, with complete CI/CD and quality processes. AI came in under spec-driven development: specifications and architecture decisions as the source of truth, not generic generation. And we laid the data foundations to treat data as a product. The result: a brand-new kindergarten-focused web application that didn't exist before, built entirely on the new architecture.