Abhijit Das — Machine Learning Researcher
About Me
I am a PhD student in Machine Learning at MBZUAI and CTO of MedOS Limited, based in Abu Dhabi. My path into this work grew from a simple conviction: if AI is to matter in science and medicine, it must earn trust where the cost of being wrong is highest. That belief shapes how I study, build, and lead. I am open to internships and visiting researcher positions — please reach out. My CV is available here.
Research Interests
- Scientific Foundation Models and World Models (pretraining, post-training, causal RL, visual CoT, test-time adaptation for multimodal LLMs, and diffusion models.)
- Autonomous Scientific Discovery for Personalized Medicine (self-evolving agents for drug discovery, wellness, and vitality.)
- AI for AI (alignment, safety, and mechanistic interpretability.)
- AI Scalability (inference optimization, edge AI models, agent-native memory, and knowledge distillation.)
What I See AI Do in Medicine in the Next 5 Years
In the next five years, AI in medicine will shift from narrow predictors to scientific partners: foundation and world models that simulate biology, agents that propose and refine personalized interventions for drug discovery and wellness, and systems that remain aligned, interpretable, and fast enough for the edge of care. The winners will not be the largest models — they will be the ones clinicians and scientists can trust to discover, explain, and act under real constraints.
All models are wrong, but some are useful.
— George Box
Updates
Joining MBZUAI for a PhD in Machine Learning.
Four papers accepted at MICCAI 2026 on OOD adaptation, hallucination mitigation in VLMs, medical image re-identification, and active visual prompting for foundation models.
Paper accepted at ECCV 2026 on hallucination mitigation for medical VQA.
Joined Prof. Imran Razzak's lab as Research Engineer I.