A principal AI engineer you can actually book.
Bring me an AI problem and I'll take it from idea to production, then hand your team a system they can run without me.
Principal-level engineering, and genuinely easy to work with: clear updates, honest trade-offs, no ego. Build, advise, or train, whatever the work needs.
Production ML since 2018
- Bosch
- ETH Zürich
- caralegal
- TrendSurfer
Ways we can work together.
Whatever shape it takes, the goal is the same: something real, in production. Happy to start small.
Build
From idea to production, end to end.
- ▸ Agent systems on the Claude stack: Skills, MCP, Cowork, Memory
- ▸ Quant and market-intelligence systems: scanners, backtests, data pipelines
- ▸ The unglamorous parts: observability, evals, deploys
Usually 2–8 weeks. Fixed scope where I can.
Advise
A senior pair of eyes when you're stuck or just starting.
- ▸ Architecture and code reviews
- ▸ AI strategy and build-versus-buy calls
- ▸ Getting a stalled project moving again
From a single call to a few days a month.
Train
Get your team fluent on the modern AI stack.
- ▸ Hands-on workshops on Claude Skills, MCP, and agents
- ▸ Patterns for shipping AI to production safely
- ▸ Built around your codebase, not slides
Half a day to a few sessions.
Embed
Drop me into your team for a stretch.
- ▸ Hands-on, in your repo and your standups
- ▸ I carry the hard parts and leave your team able to run it
- ▸ Fractional: part-time or full-time for a season
Usually 1–3 months.
Where the work shipped.
1 2025 / present
An AI-powered breakout scanner SaaS for retail traders. Under the hood the company itself runs as an agent ensemble: trading, research, coding, sales, marketing, ops, monitoring, analytics.
- Python
- · FastAPI
- · Postgres
- · Claude
- · Stripe
- · Hetzner
2 2024 / 2025
caralegal
Principal SW Engineer, GenAI & MLDevOps
Shipped the first production GenAI feature in a Legal-Tech SaaS. RAG with local and cloud LLMs, GDPR-compliant by design. I owned the AI strategy end to end.
- Langchain
- · Azure ML
- · OTC
- · Kubernetes
- · Kafka
- · Postgres
- · Metabase
3 2018 / 2023
Robert Bosch
Senior / Principal SW Engineer, Lidar Perception
A deep-learning lane-marking detector that shipped into production vehicles. Camera-lidar deep fusion. Pushed the ROS-stack throughput from 7 to 30 GBit/s.
- PyTorch
- · C++
- · ROS
- · Embedded
How I work.
- 01
Shipping beats demoing.
If it isn't taking real traffic, I don't count it as done.
- 02
Production-grade from the start.
Observability, evals, and deploy pipelines are part of the build, not a later cleanup.
- 03
You're always in the loop.
I carry the hard parts and keep you in the picture: clear written updates and decisions you can see. No surprises.
- 04
Easy to work with, on purpose.
One focused call and a short written update a week. I fit how your team already works: light on meetings, heavy on clarity.
- 05
Low-risk by default.
Fixed scope where the work allows it, T&M only where uncertainty is real. You always know what you're paying for.
- Languages
- Python ·TypeScript ·React
- AI
- Claude Skills ·MCP ·Cowork ·Memory
- Services
- FastAPI ·Postgres ·Kafka
- Infra
- Kubernetes ·Terraform ·Hetzner
- Payments
- Stripe
Engineer based in Nagold. Eight years of production ML so far: autonomous-vehicle perception at Bosch, then GenAI at caralegal as Principal SW Engineer, where I owned the AI strategy end to end. These days I'm founder of TrendSurfer, where the whole company runs on an agent ensemble. It doubles as a very honest proving ground. Research background at ETH Zürich in optical communications.
I work with a few teams at a time, so every project gets real attention. Mostly fractional and project work, with the occasional full-time role when the fit is right. If something here sounds like your problem, I'd like to hear about it.
Tell me what you're building.
Whether you've got a concrete project or you're just trying to work out if AI is even the right tool, I'd love to hear about it. No pitch required, bring the messy version. I reply within two working days, and you'll get a straight, useful answer either way.