Full-stack engineer with 6+ years building web and mobile products in TypeScript, React and Node.js, always on fully remote teams. I put LLMs into production with guardrails: model output is verified by code, every paid call is accounted for, and a human has the final say.
Available now · Remote, CET (Spain) · Full-time or contract · EU citizen
Technical owner of the AI call-analysis features in a lead-auditing platform for a US marketing agency that serves auto repair shops. Two-engineer team, and I was the largest contributor. I took the MVP from design to validated delivery in two months.
The job: turn a shop's monthly paid calls into coaching reports a client can trust, with no invented quotes and no runaway model spend.
- Model-free quote verification. Every quoted phrase is checked character by character against the transcript, then repaired from the source or dropped. No fabricated quote reaches a client report.
- LLM spend protection. Each paid model response is stored before any rule runs. Failed runs recover, and reports rebuild deterministically without new model calls.
- Batch runs of up to 300 calls (batches of 35, 8 in parallel) with SSE streaming and live progress.
- Human in the loop. A moderator decides what enters each report, with versioned decisions, vetoes, history and a "delivered" lock.
Also: deterministic call classification (voicemail, booked appointments, store hours across time zones and DST), a hash-locked prompt, multi-tenant auth, and ~31k lines of server code covered by 65 test files.
TypeScript Node.js Fastify React PostgreSQL Prisma Anthropic SDK (Claude) SSE Vitest
▶ Walkthroughs: per request only.
The code is private. A live demo is available on request.
| Project | What it is | Stack |
|---|---|---|
| rebaseline · npm | CLI to review failed Playwright screenshots in the browser and accept them as new baselines, locally or from GitHub Actions runs. No Docker. | TypeScript, Node.js, Playwright |
| aegispad-nextjs-ai · live | AI-assisted Markdown editor: revision, translation and dictation with the Gemini API. | Next.js, TypeScript, Gemini |
| hivelens | Image indexing and search for the HIVE blockchain: 9,000+ images a day with searchable metadata. | TypeScript, Node.js |
| hive-liquidity-pools-data-index | Public data index for HIVE liquidity pools: documented API, request validation, rate limiting and financial calculations. | TypeScript, Node.js |
| keychain-helper | Library that simplifies frontend integrations with the Hive Keychain wallet. | TypeScript |
| dashboard-saas | SaaS dashboard architecture demo: component design, state isolation, performance and testing strategy. | React, TypeScript |
Frontend and full-stack product engineer on a crypto wallet (Chrome extension and React Native app) used by 10,000+ people, where a bug can move someone's money.
- Built a swap widget that generated $10,000+ in direct revenue in its first 5 months.
- Set up the project's first CI and testing environment (Jest, React Testing Library, TypeScript).
- Led a mobile UI refactor that cut up to 2 seconds from critical user flows.
Code: github.com/hive-keychain
I use AI coding assistants every day, under strict guardrails, and human review is the final authority on every commit. In products, I treat the model as an expensive, fallible component: plain code checks what it says, and deterministic rules take over wherever a model is the wrong tool.
Frontend: React · TypeScript · Next.js · React Native · Vite · TanStack Query · Zustand · Tailwind
Backend and data: Node.js · Fastify · REST · PostgreSQL · Prisma · Supabase · Zod · SSE
AI and LLMs: Anthropic SDK (Claude) · prompt design · output verification · human-in-the-loop review · cost control
Testing and delivery: Vitest · Jest · React Testing Library · Playwright · CI/CD · Vercel · pnpm monorepos