Senior Data Scientist based in Stockholm, Sweden. I build ML systems that operate in the real world — not just in notebooks.

My career has been driven by one thing: working on problems where getting it right actually matters. That's taken me from bioinformatics and drug discovery, to deploying safety-critical ML in production vehicles, to now transitioning into AI safety research.

Currently at Volvo Cars — working on ML for vehicle safety analytics and LLM-based tooling for regulatory compliance (RAG, LangChain, Azure OpenAI).

Previously on the AD/ADAS team at Volvo, where I led the development of a road friction prediction model now running in production vehicles — warning real drivers about slippery roads before they lose control. Also worked on pothole detection, sensor fusion across camera, lidar and microphone data, and road damage systems feeding into Euro NCAP safety ratings.

Before that at SciLifeLab Stockholm — built mSRGAN, one of the first GANs applied to super-resolution of drug discovery microscopy images (2017, way before generative AI was a thing). The repo is pinned below.

Transitioning into AI safety research — specifically empirical work around AI control, behavioral evaluations, and scalable oversight. Years of building ML systems that fail quietly in safety-critical environments has given me a strong intuition for why evaluation and oversight are so hard in practice. I want to bring that into the field.

• 🔬 mSRGAN — First GAN for single-image super-resolution on high-content screening microscopy images (58 ⭐). Built as my master's thesis in 2017.
• 🚗 Road friction ML — Production model deployed in Volvo SPA2 vehicles for real-time slippery road detection.

skills = {
    "ML & Stats":       ["XGBoost", "LightGBM", "Random Forest", 
                         "Time-series", "A/B testing"],
    "Deep Learning":    ["PyTorch", "TensorFlow", "CNNs", "RNNs", 
                         "LSTMs", "GANs", "Sensor Fusion"],
    "LLMs & GenAI":     ["RAG", "LangChain", "FAISS", "Azure OpenAI", 
                         "Fine-tuning"],
    "MLOps & Cloud":    ["MLflow", "Docker", "CI/CD", "Azure ML", 
                         "AWS SageMaker", "Databricks"],
    "Data":             ["Python", "SQL", "PySpark", "pandas", "NumPy"],
}