AI Full-Stack Engineer. I ship AI inside real products, not demos.
The hard part of an AI feature was never the prompt. It's the product around it: orchestration, integrations, evals, and the UI/back-end that make it survive real usage. I do the whole layer.
- 🏗️ Ex-Pivot (procurement fintech): shipped the PO approval flow + NetSuite integration
- 🤖 Founder of Neige: built an autonomous AI agent platform solo, empty repo to production
- ⚡ 9 years shipping full-stack, exclusively in startups & scale-ups
- 🧰 Next.js · TypeScript · Node · FastAPI · Python · OpenAI/Anthropic APIs · agents · RAG
react-flow-auto-layout: a published npm package that auto-lays-out React Flow graphs on dagre, fixing what a plain dagre pass gets wrong on variable-size nodes: true bounding-box centering and straightened linear edges. Strict TypeScript, dual ESM/CJS, published with provenance.
ledgerloop: an end-to-end invoice procure-to-pay pipeline. A vendor PDF is read by a vision model, then 2/3-way matched, routed, and reconciled. AI sits only where it earns its keep (deriving a client's approval workflow from their HRIS, mapping org titles to signing authority then editable in plain language; reading the messy vendor document; investigating a flagged exception against unstructured records), while the money decisions are deterministic, unit-tested code, pausing for a real human decision before anything posts. Deterministic where it must be, agentic only where the trajectory is genuinely open-ended, with a live execution trace and graceful failure. Built with Mastra.
AI Invoice Parser: drop a PDF invoice, an AI agent returns a schema-validated structured object and flags anomalies. A focused take on production-grade AI: strict typing, runtime validation at every boundary, and an eval harness across 9 messy real-world formats.
🌍 Remote from Mexico City · Available for contract work · Building AI into a fintech product? Let's talk.