Oversight of Multi-Agent Clinical AI
Supervisory control for systems of interacting clinical agents: monitoring, attribution, and enforcement of safety constraints.
Malvern Madondo
Postdoctoral Associate · Duke University
I develop reinforcement learning and control methods for decision-making in healthcare, and the oversight that keeps adaptive, multi-agent AI systems within safe limits.
About
I am a Postdoctoral Associate in the Department of Surgery at Duke University, in the Kamaleswaran Lab. I work on oversight and control of multi-agent clinical AI, offline decision-making, and resource allocation in critical care.
Before Duke, I was a Research Scientist at the University of Chicago, developing patient-specific deep reinforcement learning for adaptive proton therapy. I received my PhD in Computer Science from Emory University, advised by Lars Ruthotto and supported by a Google PhD Fellowship, and studied Computer Science and Mathematics at The College of St. Scholastica.
Research
As clinical AI moves from models that predict to agents that act and adapt, I treat oversight as a control problem: specify the limits care must respect, then enforce them through a supervisory layer outside the learned model.
Health-System Resilience to Climate-Driven Disasters
Stochastic optimization and reinforcement learning for allocating hospital resources during disaster-driven surges.
Offline Learning in Critical Care & Transplantation
Safety-constrained offline decision-making from clinical data, including organ allocation and sepsis treatment.
Earlier work
-
Adaptive proton therapy planning
Patient-specific deep RL (ProtonRL) for automatic replanning in head-and-neck cancer.
-
Closed-loop neuromodulation
Learned control policies for deep brain stimulation via Hodgkin-Huxley dynamics.
-
Glycemic control for type 1 diabetes
RL and optimal control for real-time insulin delivery, including fairness across patient populations.
-
RL for sustainable agriculture
SWATGym, an open-source RL environment for crop management balancing yield and environmental impact.
Selected Publications
Full list on Google Scholar.
-
2025
Machine Learning for Health Care (MLHC), 2025
-
2025
ProtonRL: Patient-Specific Deep Reinforcement Learning Framework for Automatic Replanning in Proton Therapy for Head-and-Neck Cancer
AAPM Annual Meeting, AI in the Clinic Session, 2025
-
2025
Deep Learning-Based Applicator Selection Between Syed and T&O in High-Dose-Rate Brachytherapy for Locally Advanced Cervical Cancer: A Retrospective Study
Physics in Medicine & Biology, 2025
- 2025
-
2024
Forecasting Land-Based Environmental Variables Using Similarity Analysis and Temporal Graph Convolutional Neural Networks
US Patent Application US-20240112442-A1
-
2023
Learning Control Policies of Hodgkin-Huxley Neuronal Dynamics
Machine Learning for Health (ML4H), Findings Track, 2023
-
2023
A SWAT-based Reinforcement Learning Framework for Crop Management
AI for Social Good Workshop, AAAI, 2023
-
2023
Forecasting Soil Moisture Using Domain Inspired Temporal Graph Convolution Neural Networks
IJCAI Special Track on AI for Good, 2023
Experience
Postdoctoral Associate
Jul 2026 — PresentDuke University · Department of Surgery, Kamaleswaran Lab
Oversight and control of multi-agent clinical AI; offline decision-making and resource allocation in critical care.
Research Scientist
Jul 2024 — May 2026University of Chicago · Radiation & Cellular Oncology, Tian Lab
Deep RL for automated proton and brachytherapy treatment planning; simulated planning environments in PyTorch.
PhD, Computer Science
Aug 2019 — May 2024Emory University · Machine Learning & Inverse Problems Group
Closed-loop control for neuromodulation and glycemic control, combining neural networks with Pontryagin's Maximum Principle and HJB equations.
PhD Machine Learning & AI Intern
May — Aug 2023Eli Lilly & Company · Advanced Analytics & Data Science
Hybrid control framework for automated insulin delivery, combining neural ODEs with PID and HJB-based optimal control over Bergman minimal-model glucose dynamics.
PhD Research Intern
Summers 2021, 2022IBM Research · Responsible & Inclusive Technologies
Created and open-sourced SWATGym, an RL environment for precision agriculture adopted by 5+ research groups. Conducted research on fairness-constrained RL for diabetes management.
Software Engineering Intern
Summers 2017, 2018Meta · Crisis Response & Facebook University
Built crisis-assistance tools for Facebook's Community Help platform, increasing engagement by 39%. Co-designed an iOS app with location-based features and image recognition.
Selected Honors
- 2025Alumni Board of Directors, The College of St. Scholastica
- 2021Google PhD Fellowship, Computational Neural & Cognitive Sciences
- 2020Science ATL Communication Fellow
- 2019Webster Scholar (Honors), The College of St. Scholastica
- 2017Code2040 Tech Trek Fellow