Shaswat Patel

New York, NY

Email: shaswat178@gmail.com

I am a Machine Learning Engineer and Researcher. I hold a Master of Science in Computer Science from New York University. My research interests lie at the intersection of Large Language Models, AI Agents, and Mechanistic Interpretability.

Currently, I am a Research Scientist at New York University working with Prof. Claudio Silva on AI agents. I built Lea, an ecosystem of software that plugs into a mathematician’s existing workflow and helps them formalize mathematics in Lean. Lea wraps a Lean 4 theorem-proving agent behind a standalone proof UI and an Overleaf extension that formalizes labeled theorem blocks straight from a paper draft. I also work with Prof. He He, and previously worked with Dr. Eunsol Choi.

I’ve contributed to multiple projects focused on RAG (Retrieval-Augmented Generation) systems, fine-tuning transformer models, and leveraging machine learning to address a range of problems. Previously, I worked as a Machine Learning Engineer Intern at Studio Management, where I developed an Event Recommendation GPT system with advanced RAG capabilities — check out Outie. Before that, I was a Software Engineer at Walmart, where I developed Confluence-integrated chatbots and automated monitoring systems. I also have experience as a Machine Learning Associate at Tavlab and MIDAS LAB, where I worked on biomedical NLP tasks, visual speech recognition, and COVID gene sequencing projects.

My work has been published in venues including MICCAI MLMI Workshop and ACL Workshop. I’m also passionate about teaching and have served as a Teaching Assistant for courses in Algorithmic Problem Solving, Data Structures and Building LLM Reasoners at NYU.

selected publications

arXiv

Is Agent Code Less Maintainable Than Human Code?

Shaswat Patel, Betty Li Hou, Arun Purohit, and 4 more authors

arXiv preprint arXiv:2606.21804, 2026

arXiv Bib HTML

@article{patel2026agentcode,
  title = {Is Agent Code Less Maintainable Than Human Code?},
  author = {Patel, Shaswat and Hou, Betty Li and Purohit, Arun and Xu, Kai and Pan, Jane and He, He and Chen, Valerie},
  journal = {arXiv preprint arXiv:2606.21804},
  year = {2026},
}

EMNLP 2026

Bridging Latent Reasoning and Target-Language Generation via Retrieval-Transition Heads

Shaswat Patel and al.

In EMNLP 2026 (In submission), 2026

Bib

@inproceedings{patel2025rth,
  title = {Bridging Latent Reasoning and Target-Language Generation via Retrieval-Transition Heads},
  author = {Patel, Shaswat and et. al.},
  booktitle = {EMNLP 2026 (In submission)},
  year = {2026},
}

MICCAI MLMI

A novel momentum-based deep learning techniques for medical image classification and segmentation

In MICCAI MLMI Workshop, 2024

Bib

@inproceedings{patel2024momentum,
  title = {A novel momentum-based deep learning techniques for medical image classification and segmentation},
  author = {},
  booktitle = {MICCAI MLMI Workshop},
  year = {2024},
}

SNAM

Rumour detection on benchmark twitter datasets using graph neural networks with data augmentation

Shaswat Patel and al.

Springer Nature Social Network Analysis and Mining Journal, 2024

Bib

@article{patel2024rumour,
  title = {Rumour detection on benchmark twitter datasets using graph neural networks with data augmentation},
  author = {Patel, Shaswat and et al.},
  journal = {Springer Nature Social Network Analysis and Mining Journal},
  year = {2024},
}

Medrxiv

Bias Amplification in Intersectional Subpopulations for Clinical Phenotyping by Large Language Models

Medrxiv, 2023

Bib

ACL Workshop

A Dataset for Detecting Humor in Telugu Social Media Text

In ACL Workshop, 2022

Bib

Medrxiv

ShockModes: A Multimodal Model for Prognosticating Intensive Care Outcomes from Physician Notes and Vitals

Medrxiv, 2022

Bib

@article{patel2022shockmodes,
  title = {ShockModes: A Multimodal Model for Prognosticating Intensive Care Outcomes from Physician Notes and Vitals},
  author = {},
  journal = {Medrxiv},
  year = {2022},
}