I am a 1st year PhD student at UMD, advised by Prof. Dinesh Manocha in GAMMA Lab. I am supported by the NSF Graduate Research Fellowship. Previously, I did research in my undergrad working at MURGe Lab, where I was mentored by Prof. Mohit Bansal and Prof. Elias Stengel-Eskin.
🔥 News
- 2026.11:  Our paper “SLEuth: An Interactive Multi-Agent System for Lupus Risk Assessment” was accepted to AAAI 2027 demo track!
- 2026.11:  New preprint “ToM2Vec: Continuous Theory-of-Mind Tokens for Multi-Agent Coordination” on a new training method to improve theory-of-mind in multi-agent systems.
- 2026.08: Â Started my PhD at UMD!
- 2026.01:  Our paper “Balancing Faithfulness and Performance in Reasoning via Multi-Listener Soft Execution” was accepted to EMNLP 2026 main!
- 2026.04: Â Honored to receive the NSF Graduate Research Fellowship!
- 2026.02:  New preprint “Balancing Faithfulness and Performance in Reasoning via Multi-Listener Soft Execution” on a new training framework to improve faithfulness within LLMs.
- 2026.01:  Our paper “DART: Leveraging Multi-Agent Disagreement for Tool Recruitment in Multimodal Reasoning” was accepted to EACL 2026 main!
- 2025.12: Â Honored to receive an Honorable Mention for the CRA Outstanding Undergraduate Research Award.
- 2025.12:  New preprint “Movie Facts and Fibs (MF^2): A Benchmark for Long Movie Understanding” introducing a benchmark for narrative understanding of long open-domain movies.
📝 Publications

ToM2Vec: Continuous Theory-of-Mind Tokens for Multi-Agent Coordination
Nithin Sivakumaran, Ali Payani, Hyunji Lee, Mohit Bansal, Elias Stengel-Eskin.
- We introduce ToM2Vec, a method to train LLM agents to model their partner’s reasoning in latent space to improve coordination in multi-agent tasks.

Balancing Faithfulness and Performance in Reasoning via Multi-Listener Soft Execution
Nithin Sivakumaran, Shoubin Yu, Hyunji Lee, Yue Zhang, Ali Payani, Mohit Bansal, Elias Stengel-Eskin.
- We propose REMuL, a training framework that improves faithfulness by incentivizing a speaker model to produce reasoning that is executable by a set of listener models

Movie Facts and Fibs (MF^2): A Benchmark for Long Movie Understanding
Emmanouil Zaranis, António Farinhas, Saul Santos, Beatriz Canaverde,…Nithin Sivakumaran, et al.
- We propose MF2, a new benchmark for evaluating whether models can comprehend, consolidate, and recall key narrative information from full-length movies

DART: Leveraging Multi-Agent Disagreement for Tool Recruitment in Multimodal Reasoning
Nithin Sivakumaran, Justin Chih-Yao Chen, David Wan, Yue Zhang, Jaehong Yoon, Elias Stengel-Eskin, Mohit Bansal.
- We propose DART, a multi-agent multimodal debate framework that uses disagreement between VLM agents to address visual uncertainty

Nithin Sivakumaran, Chia-Yu Yang, Abhay Zala*, Shoubin Yu, Daeun Hong, Xiaotian Zou, Elias Stengel-Eskin, Dan Carpenter, Wookhee Min, Cindy Hmelo-Silver, Jonathan Rowe, James Lester, Mohit Bansal.
- We propose EngageVP, a new mutimodal video QA framework for stronger understanding of student engagement and behavior in classroom videos
đź’» Experience
- 2024.05 - 2024.08, NSF EngageAI Institute, Research Intern.