lars.ai
AI Developer · University Hospital Düsseldorf
Lars Müller
I build AI agents that do real work. At University Hospital Düsseldorf, I develop Monti — a system that monitors ICU patients through wearables and detects complications early.
→ currently building: an agentic AI product of my own (stealth)
Düsseldorf, DE
01
Research
Monti — early warning for cardiac patients
Continuous analysis of vital signs to detect complications early. Anomaly detection is based on self-supervised contrastive learning — currently in retrospective studies at the university hospitals in Düsseldorf and Aachen.
01
Wearables
capture vital signs directly from the patient — continuously, not point-in-time
02
Time series
are merged and prepared in real time
03
Anomaly detection
self-supervised models learn each patient's normal pattern
04
Early warning
deviations reach the clinical team before they become critical
Published in
European Heart Journal — Digital Health ↗
02
Consulting
Agents, not chatbots
I don't build RAG demos. I build agents that answer questions before they're asked — and see processes through to the end.
My current focus is on my own projects — but for the right problem, I'm reachable: book a call ↗
Agentic AI & automation
AI agents that handle multi-step business processes autonomously and traceably — from first step to finished result.
AI‑driven compliance
Agents that read and check complex documents — answering auditors' questions before they're asked.
AI in healthcare
From data capture to model — systems that survive clinical reality and regulation.
Real-time data processing
Pipelines for continuous data streams — robust under real-world conditions.
03
Projects
2025
BSI IT‑Grundschutz AI search
Precise answers to compliance questions with exact source citations — 1,000+ pages searchable.
2023
Bachelor thesis: Active Pre-Training with Phasic Policy Gradient
Unsupervised pre-training for on-policy reinforcement learning — 62.5M pre-training steps.
GitHub ↗
PDF ↗
2021
TurtleBot: autonomous navigation
Deep reinforcement learning on real hardware — sim-to-real transfer with ROS.
GitHub ↗
04
Contact