Interdisciplinary philosopher of science. Ph.D., GSN / LMU Munich; Masters from the Munich Center for Mathematical Philosophy. Dissertation: Knowledge Transfer in Cognitive Systems Theory — knowledge as control via structure mapping; using cybernetics to explain neural networks as error-controlled regulators. Cray deep-learning internship.
Spent the last year at Appy.AI on the core team developing the multi-agent harness for Violet — an AI Chief of Staff that routes specialist agents with scoped context for real workflows. Prompts, context, orchestration, steering, agent standards (skills), evals and testing.
Significant experience and experimentation with agentic tools and new models: Devin, Claude Code, Cursor, Grok Bot, Typesafe's Jev, DSPy, etc.
Projects from the SciPhi Initiative: fungi_fortress (generative game harness), fungi_chromatic (its Game Boy Color cartridge), Structural Idealism, scipfs. Co-organizer, Nederland Philosophy Club.
Agent = model + harness
Click the print to set a target. With the harness off, the model is a brain in a vat. Plan refines the goal estimate and predicts drift. Tools give live feedback for course correction. Memory keeps unfinished goals when you click somewhere new, and the verifier checks arrival.
<A>- Agent position
<a>- Self estimate
[*]- Latest goal you clicked
[g]- Estimate for the goal being pursued
[+]- Earlier unfinished goal
Shaded trails show actual paths. Dots trace estimates. Symbols sharing a cell take turns.
The dotted ring follows the agent’s current goal estimate. The latest click stays marked after arrival.
Send a task to the right specialist
Play the coordinator: match each task to a specialist’s role. Drag a task, or select a specialist from its menu. “Match next task” shows a simple automatic assignment. This is a sorting game: it checks predefined categories, without running agents or testing their work.