Research Engineer: VLA Post-Training
I’m a Research Engineer and PhD Candidate (completing Summer 2026) at the University of Surrey, specialising in reinforcement learning for distributed VLA post-training. I’m also Co-Founder and CTO of GeoSnap.
My work spans world models, Lyapunov stability functions, and sim-to-real transfer for humanoid robotics, with implementations in JAX and PyTorch. I’ve published at ICRA, NeurIPS, and IROS, and serve as a reviewer for top ML and robotics venues.
Previously, I’ve held research roles at Cambridge Consultants (humanoid RL & VLAs) and Agile Loop (agentic AI pipelines), where I led teams and delivered production research to clients including Google Cloud and Lenovo.
Research Interests
- Reinforcement learning for robotic control
- Vision-Language-Action model fine-tuning
- Neural Lyapunov functions for stability
- Continual, multi-task and meta-learning
- World models and model-based RL