I build products that get adopted, and I've spent the last three years doing it with AI agents in the loop — for workflow automation inside real products, and for the engineering process itself.
Engineering Manager at SPS Commerce in the Revenue Recovery division — and still hands-on as an IC. We automate money recovery for 1P Amazon vendors: deductions, chargebacks and disputes. The pipeline fans thousands of parallel jobs across EKS, orchestrated by Apache Airflow, with Playwright doing the extraction, Airflow parsing, and Snowflake as the load target our AI process reads to work out how to recover each dollar. Anthropic's models and Claude are part of how we build and ship it. Based in Costa Rica 🇨🇷, across distributed teams in the US, Canada and CR.
Four companies I've been part of have been acquired — two of them I co-founded:
| Appttitude | co-founder, 2014–2019 |
| Kleeen Software → Bhuma | co-founder & Head of Software Architecture, 2019–2023 → acquired by IBM (2024); the low-code platform now ships as IBM Data Apps |
| Junglytics | senior engineer, 2023–2024 → acquired by Carbon6 (2024) |
| Carbon6 | senior engineer, then engineering manager → acquired by SPS Commerce (2025), forming the new Revenue Recovery division |
Most "built with AI" repos are demos. I'm interested in the harder version: AI agents directed like a team — written constraints, critical review, and the discipline to cut a feature when the evidence says to. Outside work I keep experimenting with AI on real problems, following what the community is figuring out and putting it straight into something that has to actually work.
- reserva-training-log — a Spanish-first iPhone training log, live and free. A hobby that solved my own problem: built from three athletes' requirements, now public and used by strangers with the same one. Designed, built and operated through AI agents, and the process's most useful output was the decision to delete the AI feature the project was named after. The implementation log is the part worth reading.
- moneywiz-mcp-server — an MCP server exposing personal-finance data to AI assistants for analysis. ⭐ 12
- gmail-moneywiz-export — automated transaction extraction from email into a finance pipeline.
TypeScript · Python · Next.js / React · Node · GraphQL · Postgres · Snowflake · Airflow · EKS / Kubernetes · Playwright · MCP · Claude Code
Top skills: Apache Airflow · Data Engineering · Engineering Management
Earlier: Solidity / smart contracts / Web3.
🏊 swimmer · 👟 runner · 📚 reader · he/his