AI engineer in Berlin

I started in e-commerce frontends — product pages, checkouts, and the payment infrastructure underneath them, eventually covering 85,000+ transactions across 300+ merchants. Payments teach you quickly that a system is only as good as its worst path: the retry, the timeout, the edge case nobody wrote down. That's the habit I brought with me into AI.

Now I build AI applications end to end. The problems I find genuinely interesting sit at the seams: keeping retrieval grounded in real sources instead of plausible ones, getting a model small and fast enough to run where it's actually needed, and connecting the messy platforms a business already uses into something that runs without anyone watching it. Model quality matters, but most of the work is the system around it.

I work in the open and in small increments — a rough version running early, measured against something real, then tightened. I'd rather cut scope than ship a demo that only holds up on the happy path, and I care about the interface as much as the pipeline, because that's where people decide whether they trust the thing.

Outside of work I produce music, which is where a lot of my interest in audio and timing came from in the first place. I play basketball and swim most weeks, and I follow markets closely enough to run my own investing on data rather than instinct — the same pull toward measuring things instead of guessing.