Co-supervisors with open research projects

Find a research project

LiGHT projects are embedded in larger research programs and divided into clear subcomponents. They can support semester projects, BSc and MSc theses, internships, and longer-term research through close, collaborative supervision.

Choose a project direction that interests you, then mention the relevant supervisor(s) or project(s) in your application.

Apply to work with LiGHT Visit LiGHT's website

Mallo Henriet

Role

Full-stack Software Engineer & Designer

About

Full-stack Software Engineer with a degree in Cyber Security (EPFL-ETHZ) but also Graphic Designer! I started working on MOOVE since its beginning as a semester project.

I can work on front-end, back-end, devops, I care a lot about robustness, code quality and security.

I'm open to supervising projects in the field of software engineering, especially if they include research and development! We value a lot innovation, don't hesitate to bring your ideas to our lab!

We are coding our software in mainly Typescript/Rust/Python. You are welcome to reach out to me so that we can start developing the ideas that you can bring!

Open research projects

• MOOVE: Web application for clinical experts to evaluate and aligned models to specific context.
• Gateway: Inference orchestration layer for MOOVE, providing secure access to AI models, RAG infrastructure, and more…
• trueMOOVE: Clinical consultation tool to capture patient data and evaluate how AI-powered decision support systems can assist medical decision-making in low resources settings.

Technical interests

• User-centered design and research
• Rigorous and grounded software engineering
• Robustness and security

Fabrice Nemo

Role

Academic Coordinator

About

Engineer at LiGHT, my main interests are Software Engineering, Education and Applied Machine Learning. My current focus in Software Engineering is making software for our lab (and the world!) to help be more efficient in Machine Learning, in applying it and for education. I also coordinate the preparation of two courses for applying AI to healthcare.

Open research projects

• Software engineering for data science/education: Build or improve Python tools useful to LiGHT and other research groups, for data science or education.
• MMORE, AMMORE, GitHelp, why not your own: Extend data-processing, literature-review, and repository-question-answering tools.
• LiGHT Bootcamp: Test and improve a supervised MOOC for clinicians and computer scientists working with medical AI projects.

Technical interests

• Software engineering and data engineering
• Agents, retrieval-augmented generation, LLMs
• AI, MLOps, and education

David Sasu

Role

Technology Director

About

Interested in researching and building verifiable tools for global healthcare and humanitarian response. Current research and engineering interests include multilingual models and speech models.

Open research projects

• Multilingual clinical speech recognition
• Model architectures for humanitarian decision support
• International humanitarian law LLM

Technical interests

• LLMs, multimodal AI, and NLP
• Medical AI and global health
• Speech recognition

Trevor Brokowski

Role

Director of Innovation

About

Working on projects that push the boundary of innovation and impact for AI. I will supervise students working on applied AI projects at the intersection of health, humanitarian response, and innovation. Potential projects may involve multimodal AI, ultrasound foundation models, LLM agents, RAG systems, computer vision, clinical decision support, prototyping, evaluation frameworks, or implementation-focused AI tools.

Open research projects

• PRISM: Program for Research in Social Art and Media. Exploring AI, media, design, and social impact through creative and research-driven projects.
• LiGHT Innovation Hub: Developing a platform for innovation theory, rapid prototyping, and AI development for health and humanitarian applications.
• LiGHT Ultratron: Building a multimodal agentic foundation model for ultrasound, including image understanding, clinical reasoning, reporting, and workflow support.
• Mam-AI: Developing AI-supported tools for maternal health, obstetric ultrasound, and task-shifted care in resource-limited settings.
• Exploratory project area: Students may also work on smaller exploratory projects related to AI-enabled clinical workflows, human-centered AI design, model evaluation, data engineering, deployment, or responsible innovation frameworks.

Technical interests

• Ultrasound and multimodal AI
• Medical AI and global health
• Agentic AI, LLMs, retrieval-augmented generation, and computer vision
• Human-centered AI evaluation
• AI for Humanitarian and Low-resource Settings

Xavier Theimer-Lienhard

Role

PhD student, Meditron Lead

Open research projects

• Reinforcement Learning with Verifiable Rewards: Practice makes perfect. How far can we train models on medical exams? This project will explore Reinforcement Learning with Verifiable Rewards to improve capabilities and medical reasoning for Meditron.
• Hallucination & Over-refusal in Clinical Settings: A doctor who invents things is dangerous, and so is a doctor who says nothing. This project will explore existing evaluations for hallucination and over-refusal, and training methods to reduce them in Meditron.
• Reinforcement Learning with Rubrics from Clinician Feedback: Clinician feedback, like comments or edits, is an interesting research direction to improve AI in medicine. This project will explore use of AI generated checklists based on clinician feedback to distill the feedback in learning.
• Consultation Flow and Overconfidence: It seems advisable that models behave as good doctors, ask the questions in the right order, and do not become too confident in their diagnosis. This project asks how can we train Meditron to follow consultation flow and to not be overconfident in its answers.

Yusuf Kesmen

Role

EPFL AI Center PhD Fellow

About

PhD student working on trustworthy LLMs for high-stakes medicine. My research separates knowledge representation from inference machinery, grounding the latter in formal frameworks from statistical decision theory. Open to supervising semester projects and BSc/MSc theses on reasoning, post-training, and medical AI.

Open research projects

• Structured reasoning paradigms and agentic frameworks for reliable AI
• Post-training for reasoning and clinical alignment
• Test-time scaling and inference-time reasoning for medical decision-making

Technical interests

• LLM post-training and reasoning
• Medical AI and trustworthy AI
• Agents and retrieval-augmented generation