I build things that talk back.

I co-founded Rasa in 2016, when "chatbot" was a slur in serious engineering circles. Some of that was fair.

The bet was that conversational interfaces would eventually matter to enterprise software, that doing them well would require real machine learning (not just keyword matching), and that the right layer to build at was open source. Two of those three turned out to be correct on roughly the timeline we expected; the third took about five years longer than I'd hoped.

Ten years in, Rasa ships the conversational AI platform that sits underneath a surprising number of the phone trees, support bots, and internal assistants you've probably interacted with. The code is open, the customers are mostly banks and telcos, and the team is spread across Berlin, London, and San Francisco.

What I actually do

These days I spend about half my time on product and architecture decisions at Rasa — the kind of medium-term bets that don't show up in a sprint but shape what the platform can do in two years. The other half is split between writing (this site), speaking (conferences, mostly), and occasional advising for companies trying to get serious about conversational AI without starting from scratch.

If you've read me before, you know the themes: small models beating big ones on narrow problems, the unglamorous middle of the stack, why "agentic" is doing too much work as a word, and the deep and under-appreciated link between conversation design and UX research.

The short version

Physicist by training. Engineer by necessity. Writer because the things I want to read aren't being written by anyone else. Based in Berlin with a bike, a dog, and a strong opinion about filter coffee.