Senior Software Engineer based in Zurich with 10+ years shipping software across web, mobile, and enterprise systems. I work independently across several engagements. Currently leading development of the Flink community app at Panter AG, and building AI agent systems that run in production.
- Agentic AI: autonomous agents with real tool access, multi-agent networks, MCP servers, custom Claude Code skills/hooks/plugins, self-hosted agent runtimes
- Agent Safety: isolation enforced in code rather than in prompts, meaning read-only by construction, fixed tool interfaces, identity the model can't pass as an argument, reversible PII tokenization
- Agent-made product videos: agents that take a codebase to a narrated, branded demo film on their own, from local bootstrap with synthetic data to voice-over in any language, capture, motion design and mastering
- RAG Systems: retrieval-augmented generation, vector search, knowledge bases, AI-powered document analysis
- AI Vision Pipelines: camera streams with real-time Claude Vision analysis, intrusion detection, PDF data extraction
- AI for Business: warehouse performance analytics, supply chain optimization, construction cost control, AI-powered recommendations
- Blockchain + AI: Solana token analysis, AI classifiers for on-chain data, pgvector similarity search, real-time WebSocket trading feeds
- AI Automation: PDF-to-Excel extraction, large-scale data processing, multi-tenant SaaS platforms
| Project | Description |
|---|---|
| llm-privacy | Agent Skill for reversible PII tokenization. Masks names, e-mails and phone numbers before the model sees them, restores them in the final answer. Deterministic, multilingual, zero dependencies |
| hermes-llm-privacy | The same engine as a Hermes Agent gateway plugin, so masking and restoration happen automatically around every model call |
| ai-demo-studio | Agent skills and scripts that turn any web project, from a folder or a git URL, into a narrated, branded product film: isolated bootstrap with synthetic data, voice-over in any language (ElevenLabs, OpenAI or offline), full-quality Playwright capture, 3D device shots, animated infographics, music and mastering, produced with one command |
| Project | Description |
|---|---|
| ai-calorie-tracker | AI-powered nutrition tracker built with Next.js, Payload CMS, and Mastra AI agents |
| ai-smart-home | Smart home dashboard with live camera streams, Claude Vision intrusion analysis, and LLM-orchestrated device control (Hikvision, TapHome, Viessmann, Jablotron) |
| interioor-ai | Automated pipeline extracting manufacturing data from PDF technical drawings into Excel using Claude Vision, with 92% accuracy |
Languages: TypeScript, Python, SQL, Java, Ruby
Frontend: React, Next.js, Angular, Tailwind, Vue
Backend: Node.js, PostgreSQL, Drizzle, GraphQL, Ruby on Rails
Cloud & Infra: GCP, Vercel, AWS, Docker, Linux/systemd, GitHub Actions, Sentry
AI/ML: Claude API, Mastra, AI SDK, LangChain, LlamaIndex, PyTorch, Keras
Agentic: Claude Code skills, hooks & plugins, MCP servers, Hermes Agent, multi-agent orchestration, tool-use, evals, durable memory
Observability: Braintrust integration, tracing, logging
Data: PostgreSQL, Prisma, Drizzle, pgvector, Google Cloud SQL
MSc. in Computer Science from VSB-TUO. Former PhD researcher in 3D object recognition and pose estimation in RGB-D data. I've worked on banking products (Migros Bank), insurance platforms (Helsana), IoT/mobile apps (Schulthess), ERP systems (Dynoptic), and lab quality tools (q_alizer).
