Research

Generative world models and medical AI: models that learn how things work, from data that is expensive to label.

Publications

Surgical Vision World Model
Saurabh Koju, Saurav Bastola, Prashant Shrestha, Sanskar Amgain, Yash Raj Shrestha, Rudra P. K. Poudel, Binod Bhattarai
Workshop on Data Engineering in Medical Imaging, MICCAI, 2025
Paper PDF Code
Extractive Nepali Question Answering System
Yunika Bajracharya, Suban Shrestha, Saurav Bastola, Sanjivan Satyal
KEC Journal of Science and Engineering, Vol. 9, No. 1, pp. 95–102, 2025
DOI Code

Nov 2024 – present

–Implementing the GENIE (Generative Interactive Environments) framework for medical imaging, adapting video-trained models to healthcare.
–A synthetic data pipeline for spatial proteomics (IMC images) using diffusion models: the first successful application to complex multi-channel proteomics imaging.
–Statistical analysis of epigenetic biomarkers in cord blood from diabetic pregnancies, to identify potential autism risk factors.

Jun 2024 – Sep 2024

–RL agents from first principles (DQN, PPO, DDPG, SAC, A3C) for multi-turn strategy games.
–MCTS and Information Set MCTS for decisions under partial observability.
–Self-play training pipelines and adaptive AI inside the company's game engine.

2019 – 2024

Pulchowk Campus, Institute of Engineering, Tribhuvan University

First Division (78.3%). Ranked 43rd nationally in the entrance examination; full merit scholarship.

2019International Mathematical Olympiad: one of six students representing Nepal, Bath, UK.
2019Nepal Physics Olympiad: top 25 nationally; IPhO training selection.
2022First runner-up, Yomari CodeCamp.
2021First runner-up, Locus CodeJam.

Languages: Python, C/C++, JavaScript, SQL, MATLAB
Machine learning: RL (PPO, DQN, ISMCTS), deep learning, diffusion models, GANs, NLP
Tools: PyTorch, Lightning, W&B, Docker, Linux, SLURM, Git, Vim