Samarth P

Research Engineer in the Algorithms and Applications team at IBM Research, working on AI and quantum algorithms for solving partial differential equations.

Previously interned at Microsoft Research India, IBM Research, and CognitiveLab. Computer Science graduate from PES University, Bangalore.

I’m interested in multimodal AI, AI for science, and quantum algorithms—especially where AI and quantum computing meet. At IBM Research, I’m exploring classical and quantum methods for scientific computing, with a focus on partial differential equations.

Previously, I interned with the Advanced Reasoning team at Microsoft Research India, working on synthetic environments, reinforcement learning, and tool use for small language models. During an earlier IBM Research internship, I built agentic and retrieval-augmented tools for quantum workflows.

At CognitiveLab, I helped build Nayana, a family of document-centric vision-language systems spanning millions of synthetic pages across 22 languages. Supported by a $100,000 Meta Llama Impact Grant, this work shaped my interest in multimodal research.

My other work includes ReviewArena, an ICML 2026 AI for Science Spotlight; HyperCluster, a decentralized LLM inference system published at DAIS 2026; EarthInfer, an agentic Earth-observation framework presented at WACV 2026; and Manimator, which turns research papers into animated explanations. I’m always happy to discuss research collaborations and interesting problems—please reach out.

news

Jul 2026	Joined IBM Research as a Research Engineer in the Algorithms and Applications team, working on AI and quantum algorithms for science.
Jul 2026	Presented ReviewArena as a Spotlight paper and contributed talk at the ICML 2026 AI for Science Workshop in Seoul.
Jun 2026	HyperCluster was published at DAIS 2026, following a poster presentation at the AAAI 2026 ML4Wireless workshop.
Mar 2026	EarthInfer, our work on agentic reasoning over satellite imagery, was accepted at the WACV 2026 CV4EO Workshop.
Jan 2026	Joined the Advanced Reasoning team at Microsoft Research India to work on reinforcement learning, synthetic environments, and tool-using language models.
Oct 2025	Presented Benchmarking the Quantum Frontier at the 5th International Conference on AI-ML Systems.
Jun 2025	Joined IBM Research India as a Quantum Research Intern, working on agentic and retrieval-augmented tools for quantum-computing workflows.

selected publications

ICML Workshop
ReviewArena: A Large-Scale Cross-Conference Dataset and Benchmark for LLM Peer Review

Samarth P, Motamarri Sai Sathvik, Vyoman Jain, and 1 more author

In ICML 2026 Workshop on AI for Science, 2026

Spotlight Abs Bib HTML PDF Code

Spotlight paper and contributed talk

A dataset of 51,529 papers and 196,099 reviews across seven OpenReview venues, paired with ReviewArena-Eval, a venue-aware benchmark for evaluating LLM peer reviewers.
@inproceedings{reviewarena2026, title = {ReviewArena: A Large-Scale Cross-Conference Dataset and Benchmark for LLM Peer Review}, author = {P, Samarth and Sathvik, Motamarri Sai and Jain, Vyoman and Golugula, Shiva}, booktitle = {ICML 2026 Workshop on AI for Science}, year = {2026}, workshop = {true}, }

DAIS

HyperCluster: Decentralized Large Language Model Inference over Peer-to-Peer Wireless Networks

Samarth P, Vyoman Jain, Sanjiv Raghunandan, and 2 more authors

In IFIP International Conference on Distributed Applications and Interoperable Systems (DAIS), 2026

Also presented at the AAAI 2026 ML4Wireless workshop (non-archival).

Abs DOI Bib HTML PDF

@inproceedings{hypercluster2026,
  title = {HyperCluster: Decentralized Large Language Model Inference over Peer-to-Peer Wireless Networks},
  author = {P, Samarth and Jain, Vyoman and Raghunandan, Sanjiv and Sri, Akepati Ramya and Sharma, Richa},
  booktitle = {IFIP International Conference on Distributed Applications and Interoperable Systems (DAIS)},
  year = {2026},
  pages = {21--35},
  publisher = {Springer},
  doi = {10.1007/978-3-032-27358-1_2},
  note = {Also presented at the AAAI 2026 ML4Wireless workshop (non-archival).}
}

WACV Workshop
EarthInfer: Agentic Reasoning with Dynamic Modality Generation for Earth Observation

Samarth P and Sakshi Rajani

In WACV 2026 Workshop on Computer Vision for Earth Observation (CV4EO), 2026

Abs Bib HTML

An agentic Earth-observation framework that uses TerraMind to retrieve satellite data and dynamically generate geospatial modalities such as NDVI and land-use maps for spatiotemporal reasoning.
@inproceedings{earthinfer2026, title = {EarthInfer: Agentic Reasoning with Dynamic Modality Generation for Earth Observation}, author = {P, Samarth and Rajani, Sakshi}, booktitle = {WACV 2026 Workshop on Computer Vision for Earth Observation (CV4EO)}, year = {2026}, workshop = {true}, }

EMNLP Workshop

The Gemma Sutras: Fine-Tuning Gemma 3 for Sanskrit Sandhi Splitting

Samarth P and Sanjay Balaji Mahalingam

In Proceedings of the 9th Widening NLP Workshop (WiNLP), EMNLP 2025, 2025

Also presented at the ICML 2025 NewInML workshop (non-archival).

Abs Bib HTML PDF

@inproceedings{gemmasutras2025,
  title = {The Gemma Sutras: Fine-Tuning Gemma 3 for Sanskrit Sandhi Splitting},
  author = {P, Samarth and Mahalingam, Sanjay Balaji},
  booktitle = {Proceedings of the 9th Widening NLP Workshop (WiNLP), EMNLP 2025},
  year = {2025},
  pages = {235--241},
  workshop = {true},
  note = {Also presented at the ICML 2025 NewInML workshop (non-archival).}
}

ICCV Workshop
Nayana: A Foundation for Document-Centric Vision-Language Models via Multi-Task, Multimodal, and Multilingual Data Synthesis

Adithya S Kolavi, Samarth P, and Vyoman Jain

In ICCV 2025 Workshop on Computer Vision for Developing Countries, 2025

Also presented at the CVPR 2025 Vision Language Models for All workshop (non-archival).

Llama Impact Grant Abs Bib HTML PDF

Meta Llama Impact Grant

A synthetic document dataset with three interconnected subsets: 3 million richly annotated pages, a 22-language multilingual variant, and an information-retrieval collection of image-query pairs.
@inproceedings{nayana2025, title = {Nayana: A Foundation for Document-Centric Vision-Language Models via Multi-Task, Multimodal, and Multilingual Data Synthesis}, author = {Kolavi, Adithya S and P, Samarth and Jain, Vyoman}, booktitle = {ICCV 2025 Workshop on Computer Vision for Developing Countries}, year = {2025}, workshop = {true}, note = {Also presented at the CVPR 2025 Vision Language Models for All workshop (non-archival).} }