I'm Jan-David Stütz. I build the AI systems organizations actually run on — the agents, pipelines, and backends that do the work, and, when a person needs to stay in the picture, the interface that keeps them there. My design instincts come from years of treating every screen as a decision the user is about to make. The judgment comes from a PhD in Information Systems at Leibniz University Hannover on user-centered and neuro-symbolic AI, and from six peer-reviewed papers that are still being written. Five years of applied work — Robert Bosch GmbH, eccenca GmbH and MyActivities GmbH — sit underneath all of it.
The hard call on any AI project isn't which model to use. It's deciding where the machine should act alone and where a person belongs in the room — and being able to defend that line afterward, on cost, on efficiency, on whether anyone will adopt the thing six months in. Most AI projects don't fail on the model; they fail on that call. Drawing that line well is the service I offer, and research is how I hold the standard: every decision has to survive contact with how organizations actually absorb new tools. It's a question of judgment before it's a question of engineering, and judgment is what doctoral training was for.
At a glance
10+ projects
Shipped across autonomous agents, hybrid systems, and interface-first tools
5+ years
Shipping applied AI products
1 owner
End-to-end ownership from product strategy to production
Featured Projects
Three recent projects, one from each shape of AI work I build.

eccenca · 2026–present

Robert Bosch GmbH · 2024-2025

Robert Bosch GmbH · 2024
One mostly-autonomous agent, one hybrid, one interface-first. The set is the argument.
Who I work with
01
The diagnostic entry point. I sit with the workflow, talk to the people who live in it, and come back with a map of where automation pays, where humans should stay, and what to build first.
02
Scoped delivery when the target is defined and the path is not. One owner from first interview to production — architecture, agents, interfaces, and the joints between them, without a handoff chain.
03
Iteration on something already live. Adoption is stuck, outcomes underperform, or the human-machine split needs rethinking. I evaluate, rebuild the weak joints, and measure again.
Send the problem you're still arguing about internally — the one where nobody's sure yet which parts should run on their own and which parts shouldn't.