Rishabh Sabharwal

London, UK

I am a master’s student at the University of Edinburgh, being supervised by Prof. Jeff Pan. For my master’s thesis, I am designing a benchmark to evaluate multimodal, multi-turn Deep Research Agents (DRAs), with a focus on systematically assessing agent reasoning, tool use, and information synthesis during long research tasks.

I will be attending ICML 2026 in South Korea this July to present my work, “Multi-Turn Evaluation of Deep Research Agents Under Process-Level Feedback”. If you are attending ICML this year and want to chat, feel free to reach out via email. I am actively looking for Agentic AI and ML Engineering roles.

At the University of Edinburgh, I have taken relevant courses including Accelerated NLP, Advanced Topics in NLP, Machine Learning Systems, and Probabilistic Modelling and Reasoning. Through these courses, I have worked on several projects such as post-training of small-scale LLMs, Agentic RAG-based systems, optimizing an ASR system via GPU programming in Triton, and a FastAPI Safety Gateway for local LLMs. Please refer to my resume for further details on my lastest projects and work experience.

Research keywords

Multi-Agent Systems · LLM Agents · Agentic Workflows · Post-Training of LLMs

news

Jun 08, 2026	Preliminary work of my thesis has been accepted for an oral presentation (top-6%) at the SCALE Workshop, ICML 2026!
Apr 30, 2026	Paper accepted to the C3NLP Workshop, ACL 2026.
Jan 16, 2026	Paper accepted at IEEE Transactions on Artificial Intelligence (Q1 Journal).
Sep 15, 2025	Started my MSc in Artificial Intelligence at the University of Edinburgh.
Jul 31, 2025	Completed my Predoctoral Research role at IISc Bengaluru (VISTA Lab).

selected publications

ICML

Multi-Turn Evaluation of Deep Research Agents Under Process-Level Feedback

Rishabh Sabharwal, Hongru Wang, Amos Storkey, and 1 more author

In SCALE Workshop, ICML 2026, 2026

Bib HTML

@inproceedings{sabharwal2026multiturnevaluationdeepresearch,
  title = {Multi-Turn Evaluation of Deep Research Agents Under Process-Level Feedback},
  author = {Sabharwal, Rishabh and Wang, Hongru and Storkey, Amos and Pan, Jeff Z.},
  booktitle = {SCALE Workshop, ICML 2026},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {cs.AI},
  url = {https://arxiv.org/abs/2606.09748},
}

ACL

Does Reasoning Kill the Joke? Long-Context Humor Understanding in Hindi

K. Anuranjana, N. Shrivastava, Rishabh Sabharwal, and 4 more authors

In C3NLP Workshop, ACL 2026, 2026

Accepted; to appear

IEEE TAI

On Convergence of Adam with Data Dependent Stepsize

A. Mazumder, Rishabh Sabharwal, M. Tayal, and 3 more authors

IEEE Transactions on Artificial Intelligence, 2026

Bib HTML

@article{mazumder_adam_stepsize_tai,
  title = {On Convergence of Adam with Data Dependent Stepsize},
  author = {Mazumder, A. and Sabharwal, Rishabh and Tayal, M. and Kumar, B. and Garg, C. and Rathore, Punit},
  journal = {IEEE Transactions on Artificial Intelligence},
  year = {2026},
}