Machine-learning engineer; Basel, Switzerland

ML & AI engineer, focused on speech and data.

I work on Swiss-German ASR, retrieval-augmented LLMs, and the datasets and pipelines behind them; at the FHNW Institute for Data Science.

What I'm working on ↓

Now — 2026

What I'm working on

Most of my time goes to Swiss-German speech — teaching models to hear and speak a language with almost no data; together with Swiss companies putting it into production.

Nemotron 3.5 ASR

Fine-tuning NVIDIA's Nemotron ASR for Swiss German; pushing accuracy and latency for real-time transcription.

Whisper fine-tuning

Production Swiss-German ASR, fine-tuned and deployed to more than a million users. Read the paper ↗

Qwen3-TTS

Swiss-German dialect voices via talker-layer fine-tuning; expressive speech from a model with no dialect data. Hear it →

Real-time STT

Working with Swiss-based companies to bring low-latency, real-time speech-to-text into production settings.

Data preparation

Building the pipelines that turn raw, messy audio into clean training data; alignment, segmentation, and quality filtering at scale.

Evaluation

Designing benchmarks and WER metrics that actually reflect dialect performance; so model gains hold up outside the lab.

Open work

Open-sourcing datasets on Hugging Face, building space-weather models for solar radio bursts, and competing in the Numerai tournament.

Where to next

Read the writing → Browse the projects → Get in touch →