I build products end-to-end — from React/Next.js front-ends to Node/Python back-ends — and I'm now bringing AI into production.
- 🏦 At MODO (Argentine fintech) — building AI-powered features with real tests and guardrails.
- 🐾 Building Firulapp — a community app for pet owners in Buenos Aires: social feed + AI-powered Lost & Found. Next.js 16, React 19, Firebase, hexagonal architecture.
- 📚 14+ years shipping software. TDD, clean architecture, and pragmatism over perfection.
Capabilities I build with LLMs (framed as reusable skills — no proprietary detail):
- LLM agents & tool use — agents that call tools reliably, with structured outputs.
- MCP servers — exposing application capabilities to AI assistants via the Model Context Protocol.
- RAG — retrieval-augmented generation grounded on domain data.
- Prompt engineering & evals — versioned prompts, structured outputs, automated evaluation.
- AI safety & guardrails — prompt-injection defenses, access control on AI surfaces, output validation, and parity checks that keep AI features in sync with the product.
- 🐾 Firulapp — Pet community + AI Lost & Found · Next.js · Firebase · LLM agents
- 🧰 nestjs-api-tools — NestJS toolkit with a TOON interceptor that cuts LLM token usage 30–60% vs JSON
- 💳 stripe-plans-importer-nodejs — Programmatically import Stripe plans in Node.js
- 🛒 tiendarapida — MVP de tienda online self-hosteable · catálogo en Google Sheets + pago con MercadoPago/MODO · Next.js
- 🧪 monopoly-tdd — Test-driven domain modeling in TypeScript (Jest)
- 🎯 valorolette — Valorant agent/map roulette · React + Sass (fan project, just for fun)
Open to interesting problems in AI engineering & full-stack product work.