// I want to understand how the world works.
I am a junior in Computer Science at the Indian Institute of Technology Delhi, broadly interested in machine learning, from LLM reasoning and its failure modes to molecular property prediction. I am at the University of Waterloo as an exchange student this Fall, and I'm an incoming Quantitative Research Intern at IMC Trading for the upcoming summer.
Previously, I was a Research Assistant at CLAN, Aarhus University with Prof. Akhil Arora, studying LLM understanding and reasoning, error correction, and hallucination mitigation via graph learning. I co-authored ReasonBENCH, an open-source benchmark of the variance and (in)stability of LLM reasoning. This visit is supported by a Danish Data Science Academy research grant of 15,000 DKK.
Previously, I worked at AIMES & DSIRe, IIT Delhi with Prof. Tarak Karmakar and Prof. Sayan Ranu, where I developed Dissolvr, an interpretable solubility prediction framework (ICML 2026), and SC3, a multi-solvent solubility benchmark with 101k+ measurements.
International Conference on Machine Learning (ICML) 2026, Seoul
A physically-constrained gradient-boosted framework using descriptor-based representations, solute-solvent interactions, and LLM-assisted explanations.
paper ·
arXiv preprint, 2026
A multi-solvent benchmark of 101k+ measurements estimating an aleatoric floor of ε = 0.106 log S; evaluated 31 models across six families.
paper ·
EIML Workshop @ ICML 2026
Measures variance across 12 models, 10 reasoning strategies, and 30 repeats per configuration; establishes a two-component noise taxonomy.
paper ·
I served as Convenor of the Debating Society at IIT Delhi, used to write for the Board for Student Publications (Top Contributor, June 2025), and coordinate ML research at ARIES, the AI/ML club. For honors, coursework, projects, and skills, see my academics page; course notes and tutorials live on the resources page.