Dheeraj Vattikonda

Visiting Researcher, ServiceNow AI Research
M.Sc. Student, McGill University & Mila

Google Scholar / 𝕏 @dheeraj_46329 / LinkedIn

About Me

I work on reinforcement learning and reasoning with LLM agents in the Long Horizon Agents team under Massimo Caccia at ServiceNow AI Research, based out of Montreal. I am also a Master's student at McGill University and Mila under the guidance of Xue (Steve) Liu.

My current research revolves around reasoning in LLM web agents and tool-calling systems. My recent work on web agent training received an oral presentation at ICML and will appear at NeurIPS 2025.

I completed my Bachelor's in Electronics and Communication Engineering at NIT Hamirpur, where I worked on robot perception, focusing on differentiable SLAM systems and LiDAR-based perception tasks for autonomous navigation.

News

Feb 2026 New paper: π-Distill — Privileged Information Distillation for Language Models
2025 Oral presentation at ICML 2025 CUA Workshop
2025 "How to Train Your LLM Web Agent" accepted at NeurIPS 2025
2024 Aurora accepted as Spotlight at NeurIPS 2024
Aug 2024 Started as Visiting Researcher at ServiceNow AI Research
Aug 2023 Started M.Sc. at Mila / McGill University

Selected Publications

Privileged Information Distillation for Language Models

Emiliano Peñaloza, Dheeraj Vattikonda, Nicolas Gontier, Alexandre Lacoste, Laurent Charlin, Massimo Caccia

arXiv 2026

Distill frontier models even when they hide their reasoning, using training-time privileged information.

paper code

How to Train Your LLM Web Agent: A Statistical Diagnosis

Dheeraj Vattikonda, Santhoshi Ravichandran, Emiliano Penaloza, Hadi Nekoei, Megh Thakkar, Thibault Le Sellier de Chezelles, Nicolas Gontier, Miguel Muñoz-Mármol, Sahar Omidi Shayegan, Stefania Raimondo, Xue Liu, Alexandre Drouin, Laurent Charlin, Alexandre Piché, Alexandre Lacoste, Massimo Caccia

ICML 2025 CUA Workshop (Oral) · NeurIPS 2025

A comprehensive statistical analysis of training recipes for LLM web agents.

paper slides