PhD candidate in Generative AI at the University of Stuttgart and the Max Planck Institute for Intelligent Systems, Germany.
I develop efficient and symmetry-aware generative models, with a focus on diffusion models, flow matching, equivariant learning, and scientific machine learning. My goal is to apply these methods to molecular modelling, chemistry, biology, and scientific discovery.
I expect to submit my PhD thesis in February 2027.
Website · Google Scholar · LinkedIn · Email
- Diffusion models and efficient generative inference
- Flow matching and one-step generative models
- Equivariant and symmetry-aware machine learning
- Generative modelling for molecules, crystals, and chemical reactions
- Scientific machine learning and AI for health
I introduced LD3, a lightweight approach for learning sampler-specific diffusion time discretizations. It improves few-step sampling quality for pre-trained diffusion models without retraining the base model, and was evaluated across pixel-space and latent-space models.
Co-first-author work on symmetry-aware one-step generative modelling. SymDrift uses optimal alignment or invariant embeddings to enable efficient generation for highly symmetric distributions, including molecular conformers and transition states.
Generative flow-matching approach for refining low-fidelity transition-state structures. The method improves transition-state localisation and accelerates high-level quantum optimisation.
During my Applied ML Research internship at Black Forest Labs, I contributed to Self-Flow, a self-supervised flow-matching framework for scalable multi-modal synthesis. My work studied the semantic information learned by latent embedding layers after training.
I was an Applied ML Research Intern at Black Forest Labs from October 2025 to April 2026.
There, I applied research on diffusion-ODE time-step optimisation to develop an improved sampling schedule for Flux 2, supporting high-quality generation with few sampling steps across image resolutions.
Flux 2 sampling implementation
Before my PhD, I worked on graph neural networks and knowledge-graph completion at VinAI Research. This work resulted in publications at ESWC and Findings of EMNLP.
- Reviewer: ICLR, ICML, NeurIPS, EMNLP, ECCV, and WACV
- Teaching Assistant: Introduction to Artificial Intelligence and Reinforcement Learning
- Mentor for Bachelor's and Master's thesis projects