Most of what I do professionally comes down to one question: what do we actually know, and how confident should we be? My academic research centered on building uncertainty quantification into deep learning—developing generative models that don’t just give you an answer, but tell you how much to trust it and where it might break. That mindset carries into how I consult: I’d rather give you an honest “this probably won’t work” early than let you discover this six months later.
Academic background
I have a PhD in machine learning from Heidelberg University (summa cum laude), where I developed deep generative models for inverse problems—situations where you’re trying to infer causes from effects, and where there’s often more than one plausible explanation. I was awarded the Ruprecht-Karls-Prize for my thesis on conditional invertible neural networks.
Before that, I studied physics (BSc and MSc, both at Heidelberg), which instilled in me the habit of caring whether models actually reflect reality rather than just fitting training data.
The methods I developed found applications across a range of fields: astrophysics, medical imaging, particle physics, and mechanical engineering. Working with domain experts taught me to translate between ML abstractions and real-world constraints.
I’ve accumulated around 2,500 citations and an h-index of 19. I also co-created FrEIA, an open-source framework for invertible neural networks that’s still in active use.
Industry background
From 2022 to 2025, I was Head of Machine Learning at Copresence, a startup building 3D avatar reconstruction from smartphone video. I joined as a founding technical team member and contributed to geometry reconstruction, appearance modeling, and the production pipeline. The system outputs rigged meshes with PBR textures, exportable to Blender, Unreal, Unity, and other common 3D software.
What I took away from this role: uncertainty quantification and outlier detection aren’t academic luxuries. The pipeline works because we built robustness and failure detection into every step. That experience shapes how I think about feasibility when advising others.
How I work
I take on focused, time-boxed engagements rather than ongoing integrations. A typical project might be a feasibility study, a technical audit, a prototyping sprint, or a consulting retainer for decision support.
I’m not a production engineer. If you need someone embedded in your team writing and maintaining code long-term, I’m not the right fit. If you need someone to assess whether an approach will work before you commit, review a technical direction, or prototype something quickly to inform a decision, that’s where I’m useful.