Engineer turned regulated-AI program leader. Building privacy-first AI in pharma at Roche — and indie iOS apps that run AI on-device. Based in Basel.
-
Ahimo — privacy-first iOS health app. Pairs Apple Watch + HealthKit data with the context only you can provide (mood, stress, drinks, symptoms) and uses on-device LLMs (Apple Foundation Models or MLX: SmolLM3, Qwen, Llama) to surface real personal patterns. No cloud processing of health data. Public docs →
-
Falcata Run — Apple Watch-first sprint training app. Track reaction time, drive force, cadence, stride length, and gait symmetry at 100 Hz. Built for sprinters, coaches, and Masters athletes. Public docs →
Five years at Roche leading programs at the intersection of regulation, digital health, and AI:
- Agentic AI in regulated pharma — building privacy-first RAG agents under a hard no-hallucination constraint. Every claim has to trace back to its source; if a fact can't be grounded in retrieved evidence, it gets discarded rather than paraphrased. That bar is high but it's the only way agentic systems are usable in pharma. Backed by MIT Professional Education's Applied Agentic AI for Organizational Transformation (Feb 2026).
- Regulatory leadership — IVDR adoption and MDR compliance across the diagnostics portfolio; remediation with BfArM and SwissMedic; rollout of a regulatory framework across 27 EU + 4 EFTA countries.
- Digital endpoints — design and validation of novel clinical endpoints from wearable devices.
- Federated research in multiple sclerosis — multi-center studies with universities and clinics, shaping both technical infrastructure and legal frameworks for secure data sharing.
- Regulatory information lifecycle — Veeva RIM integration, automated content generation, terminology standardization across the submission pipeline.
- Brain in the Loop — my blog on AI alignment, agentic workflows, and privacy-preserving health technology. Also on Substack.
- On-Device SLM Investigation — how four 3B-class on-device models (SmolLM3, Qwen3.5, Llama-3.2, Phi-3.5) actually behave on HealthKit-style correlation prompts and agentic health-logging tool calls.
- MIT Professional Education — Applied Agentic AI for Organizational Transformation (2026)
- ETH Zürich — CAS Cyber Security (2019–2020)
- Stanford — Energy Innovation and Emerging Technologies Certificate (2012–2013)
- M.Sc. Applied Mechanics + B.Sc. Robotics — Technical University of Ostrava
- 20 years across pharma, automotive, nuclear, and energy — Roche, Ansaldo Energia (gas turbines), Siemens Energy (nuclear steam turbines), Cummins, Honeywell, ŠKODA
- Other certifications: Six Sigma Green Belt, PSM I, CSPO
- Languages: Czech (native), English (fluent), German (working), Spanish (basic)
- iOS / Swift / Rust / Python / on-device ML