Jishnu S Nair

Senior ML Engineer at ServiceNow

I am a Senior Machine Learning Engineer at ServiceNow, working on reasoning in large language models and reinforcement learning for agentic systems. Most of my time goes into figuring out how to make LLMs better at complex decision-making and how multiple agents can work together on real-world enterprise problems.

Broadly, I’m interested in training reasoning models, reinforcement learning for LLMs, synthetic data generation, building evaluation benchmarks for generative AI, and agentic systems.

Before this, I spent a few years building semantic search, enterprise-level clustering services, and NLP infrastructure at ServiceNow.

I did my undergrad in Computer Science at Government Model Engineering College, Kochi.

For updated details, see my Google Scholar / LinkedIn / GitHub pages.

selected publications

ICML
Do Enterprise Systems Need Learned World Models? The Importance of Context to Infer Dynamics

Jishnu Sethumadhavan Nair, Patrice Bechard, Rishabh Maheshwary, Surajit Dasgupta, and 13 more authors

In ICML 2026 Workshop on Failure Modes in Agentic AI, Jul 2026

Abs arXiv Bib

World models enable agents to anticipate the effects of their actions by internalizing environment dynamics. In enterprise systems, however, these dynamics are often defined by tenant-specific business logic that varies across deployments and evolves over time, making models trained on historical transitions brittle under deployment shift. We ask a question the world-models literature has not addressed: when the rules can be read at inference time, does an agent still need to learn them? We argue, and demonstrate empirically, that in settings where transition dynamics are configurable and readable, runtime discovery complements offline training by grounding predictions in the active system instance. We propose enterprise discovery agents, which recover relevant transition dynamics at runtime by reading the system’s configuration rather than relying solely on internalized representations. We introduce CascadeBench, a reasoning-focused benchmark for enterprise cascade prediction that adopts the evaluation methodology of World of Workflows on diverse synthetic environments, and use it together with deployment-shift evaluation to show that offline-trained world models can perform well in-distribution but degrade as dynamics change, whereas discovery-based agents are more robust under shift by grounding their predictions in the current instance. Our findings suggest that, in configurable enterprise environments, agents should not rely solely on fixed internalized dynamics, but should incorporate mechanisms for discovering relevant transition logic at runtime.
@inproceedings{nair2026enterprisesystemsneedlearned, title = {Do Enterprise Systems Need Learned World Models? The Importance of Context to Infer Dynamics}, author = {Sethumadhavan Nair, Jishnu and Bechard, Patrice and Maheshwary, Rishabh and Dasgupta, Surajit and Ramachandran, Sravan and Bhagat, Aakash and Radhakrishna, Shruthan and Pattnaik, Pulkit and Obando-Ceron, Johan and Malay, Shiva Krishna Reddy and Davasam, Sagar and Subramanian, Seganrasan and Mittal, Vipul and Nemala, Sridhar Krishna and Pal, Christopher and Sunkara, Srinivas and Rajeswar, Sai}, year = {2026}, month = jul, eprint = {2605.12178}, archiveprefix = {arXiv}, primaryclass = {cs.AI}, booktitle = {ICML 2026 Workshop on Failure Modes in Agentic AI}, }
ICML
EnterpriseOps-Gym: Environments and Evaluations for Stateful Agentic Planning and Tool Use in Enterprise Settings

Shiva Krishna Reddy Malay, Shravan Nayak, Jishnu Sethumadhavan Nair, Sagar Davasam, and 5 more authors

International Conference on Machine Learning (ICML), Mar 2026

Abs arXiv Bib

EnterpriseOps-Gym is a large-scale benchmark for evaluating the agentic planning and tool-use capabilities of LLM agents across enterprise operations. It comprises 1,150 expert-curated tasks spanning 8 enterprise domains, each running against live containerized MCP servers backed by realistic, fully synthetic databases. Unlike static QA benchmarks, EnterpriseOps-Gym evaluates agents on final environment state using SQL verifiers - meaning agents are rewarded for achieving the correct outcome, not for following a rigid action sequence. Tasks require long-horizon multi-step reasoning, strict policy compliance, and precise tool invocation under complex data dependencies.
@article{malay2026enterpriseopsgymenvironmentsevaluationsstateful, title = {EnterpriseOps-Gym: Environments and Evaluations for Stateful Agentic Planning and Tool Use in Enterprise Settings}, author = {Malay, Shiva Krishna Reddy and Nayak, Shravan and Sethumadhavan Nair, Jishnu and Davasam, Sagar and Tiwari, Aman and Madhusudhan, Sathwik Tejaswi and Nemala, Sridhar Krishna and Sunkara, Srinivas and Rajeswar, Sai}, journal = {International Conference on Machine Learning (ICML)}, year = {2026}, month = mar, eprint = {2603.13594}, archiveprefix = {arXiv}, primaryclass = {cs.AI}, }

AAAI

DNR Bench: Benchmarking Over-Reasoning in Reasoning LLMs

Masoud Hashemi, Oluwanifemi Bamgbose, Sathwik Tejaswi Madhusudhan, Jishnu Sethumadhavan Nair, and 2 more authors

AAAI Conference on Artificial Intelligence 2026, Mar 2025

Abs arXiv Bib

@article{dnrbench2025,
  title = {DNR Bench: Benchmarking Over-Reasoning in Reasoning LLMs},
  author = {Hashemi, Masoud and Bamgbose, Oluwanifemi and Madhusudhan, Sathwik Tejaswi and Sethumadhavan Nair, Jishnu and Tiwari, Aman and Yadav, Vikas},
  journal = {AAAI Conference on Artificial Intelligence 2026},
  year = {2025},
}