Applying AI to push boundaries.
CTO based in Berlin. I build things at the intersection of AI and product — autonomous agents that manage engineering teams, voice bots that handle real phone calls, developer tools that make Claude Code actually useful in production.
I ship fast, open source a lot, and believe the best software is built by small teams with high leverage.
Autonomous AI Systems — Most of my recent work is about making AI agents that actually do real work, not demos. Zoro is an autonomous engineering manager that runs as an iTerm2 extension — it monitors tickets, routes work to Claude Code sessions, detects error loops, and runs a web cockpit for oversight. claude-bootstrap (529+ stars) is the opinionated project scaffold I use to make Claude Code reliable across all my projects.
AI-Native Developer Tools — Hive is a standalone AI command center for SaaS — it manages budgets, creates tasks, makes strategic decisions, and coordinates between AI agents and humans. Halo brings Claude Code to the desktop. voxy is a voice-controlled terminal assistant.
Voice & Conversational AI — AIVoiceBot is a complete voice bot service handling inbound and outbound calls. realtime-transcription does live audio-to-text across languages. voiceover generates AI narrations for screen recordings.
MCP & Integrations — mcp-linkedin-server (50+ stars) is an MCP server for LinkedIn automation. I've built crawlers, search engines, proposal generators, and various connectors between AI and the tools people already use.
Alongside open source, I lead engineering at three companies:
Enterprise CX Platform — CTO of a customer experience platform processing millions of survey responses. I manage the full stack: backend services, frontend apps, Shopify integrations, and a migration from legacy to a modern multi-tenant architecture. Multi-provider integrations (Salesforce, HubSpot, Intercom). Team of engineers across backend, frontend, and DevOps.
End-to-End AI Marketing Agents — CTO of an AI-native marketing platform built around a fully autonomous AI agent that controls the entire system. The agent runs campaigns end-to-end: strategy, brief co-creation, content generation, creative production, and analytics — all via streaming chat, autopilot mode, or proactive nudges. It conducts interviews and outreach over email, WhatsApp (@-mention driven), and live voice meetings with real-time turn-taking and TTS. Multi-agent architecture with hired sub-agents per brand, automated email cadences, MCP integrations (HubSpot, Salesforce, Google Ads, social platforms), and a full copilot toolkit with tools, skills, and context-aware planning.
AI-Powered Learning & Transformation Platform — Technical co-founder of an ed-tech platform with a suite of AI products: synthetic interviews, knowledge sprints, AI-generated podcasts, micro-learning, chat-based tutoring, content generation, onboarding companions, and a full transformation suite for organizational change. Simulation agents, ITSM process automation, and a unified platform serving it all.
I wrote a No-Agile Agile Manifesto and an Organization Consciousness Protocol because I think most process is theater. What actually works:
- Small teams, high autonomy. One engineer with good tools beats a squad with a Jira board.
- AI as a multiplier, not a replacement. I use Claude Code for ~80% of my implementation. The other 20% is architecture, judgment, and taste.
- Ship first, abstract later. Three similar lines of code is better than a premature abstraction.
- TDD when it matters. Tests are a design tool, not a checkbox. I write them first for complex logic, skip them for throwaway scripts.
Python TypeScript FastAPI Claude API Agent SDK MCP SQLite PostgreSQL React Next.js Shopify iTerm2 API WebSocket GraphQL