I build full-stack AI products end to end and run them in production.
Right now that's PracticeDepth: an adaptive AI technical interviewer I designed, built and operate on my own. It's been live since July 2026 and taking real payments since August.
- PracticeDepth: a LangGraph.js state machine that interviews like a senior engineer, with voice in and out (Whisper STT, Gemini TTS), code challenges, CV defense and a RAG-grounded support assistant. Now adding a B2B screening product for hiring teams.
- How I ship it: tickets in Linear run as parallel background Claude Code agents on GitHub Actions. Each one opens a CI-checked draft PR (1,100+ tests). I review every change before it merges, and payment code is kept human-only by a CI guard.
- WohnIQ: AI apartment search for Berlin's rental market, with pgvector semantic search and an explainable scam-risk score.
- Product ownership end to end: scoping, UI, APIs, data, payments, deployment, monitoring
- AI in production: agent state machines, speech in and out, cost-aware model choices, grounded assistants, MCP
- Real-time and event-driven systems: WebRTC, WebSockets, Kafka, MQTT, from 3+ years on EU emergency-response projects
- Shipping safely: feature flags, CI gates, written architecture decisions (ADRs), self-hosted ops with Sentry and Grafana
- Frontend: React · TypeScript · Next.js · Tailwind
- Backend: Node.js · NestJS · Python · FastAPI · PostgreSQL · Prisma · Redis
- AI: Claude Code · LangGraph · Gemini · pgvector · MCP
- Infra: Docker · GitHub Actions · Coolify · Hetzner · Sentry · Linear
Open to AI-native product engineering roles and interesting collaborations.
✉️ martin@martinoputa.com · 🌐 martinoputa.com · 💼 LinkedIn
