Johannes Moll

Currently I am a PhD candidate at the Technical University of Munich supervised by Daniel Rueckert and advised by Keno Bressem, Lisa Adams, and Jiazhen Pan.

My research includes two major threads:

Building multi-turn, tool-using agents that can reason over longitudinal patient records, retrieve relevant evidence, and produce structured, citation-grounded clinical answers.
Improving model accuracy and interpretability with chain-of-thought traces and faithfulness evaluation.

bio

Prior to beginning his PhD, Johannes spent six months as a research assistant at the Artificial Intelligence in Medicine & Imaging (AIMI) center at Stanford University, where he was supervised by Curtis Langlotz and Jean-Benoit Delbrouck.

Aug 2026	Our paper GRASP: Gated Regression-Aware Skill Proposer for Self-Improving LLM Agents was accepted to the EMNLP 2026 main conference in Budapest, Hungary.
May 2026	Released GRASP, a self-improvement method for LLM agents that gates every skill-library edit on a regression-aware probe. Joint work with Microsoft Healthcare & Life Sciences. Paper, code.
Apr 2026	New preprint out: Agentic clinical reasoning over longitudinal myeloma records.
Oct 2025	Our paper on Evaluating Reasoning Faithfulness in Medical Vision-Language Models using Multimodal Perturbations was accepted to the ML4H symposium in San Diego, CA, USA.
Aug 2025	Our paper on Structuring Radiology Reports: Challenging LLMs with Lightweight Models was accepted to the EMNLP2025 main conference in Suzhou, China.