Blog

Notes on reliable AI.

LLM reliability, hallucination mitigation, agentic system design, and what actually works when AI meets regulated industries.

July 15, 2026·3 min read

Prompt Risk Density: measuring hallucination risk before the model answers

Most hallucination research fixes the model. My bachelor's thesis, Echo, asks a different question: how much of the risk is already sitting in the prompt, and can we measure it before generation starts?

LLM ReliabilityHallucination MitigationRead post

July 15, 2026·3 min read

Consensus without an orchestrator: what building Mycellium taught me

Most multi-agent systems put a coordinator in charge, and the coordinator becomes the bottleneck. In the TUM × SAP practical course we built Mycellium the other way around: no orchestrator, just an auction, a blackboard, and a consensus rule.

Agentic System DesignMulti-Agent SystemsRead post

Prefer a feed? Subscribe via RSS.