Research Intern, NAAMII
I am currently a Research Intern at NAAMII, working with Dr. Prashnna Gyawali on the reliability and trustworthiness of large language models in healthcare. My interests are broad, from reinforcement learning to applications of machine learning in science, and I enjoy exploring new directions.
I completed a B.Sc. in Computational Mathematics at Kathmandu University in 2026.
I am looking for PhD opportunities starting in Spring/Fall 2027.
In progress · 2026 – present
RL with verifiable rewards checks only the final answer, never the reasoning. I am showing that, because each update is so noisy, training can permanently lock in wrong reasoning even when correct reasoning is strictly better, and I derive the probability of this in closed form.
Under review · ML4H 2026
Extends distribution-free risk guarantees to tier-specific risk budgets, so clinically dangerous claims are held to a stricter bound than benign ones. The guarantee holds for any scoring function.
Under review · MLHC 2026
Conformal Claim Refinement (CCR) splits an LLM response into atomic claims, filters high-risk ones with claim-level conformal risk control, and rewrites the answer. It cuts sycophancy by up to 7.4% on average with no retraining.
Nepal Applied Mathematics and Informatics Institute for Research (NAAMII)
Computational Mathematics Club, Kathmandu University
