Open to remote roles & freelance
Full-stack AI engineer in Spain. Web products designed, built and operated end-to-end: live Stripe billing with paying subscribers, a validated integration with Spain's tax agency, and AI features people pay for — built by orchestrating Claude Code, run with production discipline.
Selected work
Every project below is deployed and in use — the important ones bill money or talk to a government API. Each covers the full cycle: product, code, data model, deployment, operations.
SaaS for freelance dentists working across clinics: log daily clinical production, see earnings analytics, and photograph a paper production sheet so the AI scanner (Claude vision) extracts every line item. Live Stripe billing at €12/mo · €99/yr, with real subscribers.
Invoice-integrity engine for Spain's VeriFactu anti-fraud law (mandatory for healthcare providers by mid-2027). Each invoice gets a SHA-256 fingerprint chained to the previous one, a verification QR, and is submitted as signed XML to the AEAT tax agency over SOAP on a cron schedule — validated end-to-end against the government's test environment.
White-label appointment-booking bot engine. One deployment serves many clinics: per-clinic configuration from an admin panel, deterministic conversation flows (no hallucination risk on medical scheduling), and transactional email for confirmations.
An AI agent that manages Google Ads for a real dental clinic. Every day it reads performance data and decides — stop, scale, optimize or explore — under a hard spend cap, optimizing for phone calls, then reports to the non-technical owner in plain language. Case study on request.
Clinic-management platform: appointment scheduling, patient records with an interactive FDI dental chart, and VAT-exempt healthcare invoicing built to Spanish rules.
Time-tracking for a real company: GPS-geofenced clock-in, payroll summaries, vacation workflows and role-based auth built from scratch. Compliant with Spain's mandatory time-registry rules. Demo on request.
Also built
Working method
Claude Code agents run the full lifecycle — spec, build, test, deploy. The AI produces the volume; engineering judgment owns correctness: verification against the live product, security review, data modeling, and the final call on what ships.
Nothing counts until it runs live with real users. Small releases, verified in production, with critical paths — billing, patient data, tax records — defensive by default: secrets in environment variables, parameterized queries, graceful failure for everything secondary.
Written-first communication, Spanish native, working English, UTC+1 with overlap for US mornings. Deep domain in healthcare and dental operations plus Spanish regulatory compliance — built for real clinics, not hypothetical users.
Stack
Four working demos — an AI assistant, a binary file parser, a data extractor and a small game engine — each running live in your browser, right here. Nothing below is a screenshot.
Describe a product or a problem. A Claude model streams back what it would build, the stack it would use, and one risk worth flagging — token by token, in real time.
Upload an activity file from a sports watch. The raw binary is decoded on the server into a GPS track, speed and elevation — then drawn from scratch as SVG.
Paste an email or a note. The model returns strict, schema-validated JSON — people, organizations, dates, action items — the backbone of any real automation.
A tiny reflex game, hand-built on an HTML canvas with its own game loop and state. Sometimes the fastest way to show you can build is to let someone play.
Tap the ring (or press Space) the moment the marker crosses the teal zone. It only gets faster.
Every demo above runs live — frontend, backend, AI and all.
built with Next.js · TypeScript · Tailwind · the Claude API