P Sam Sahil

P Sam Sahil

Visiting UG Research Fellow

Harvard University

Biography

I am an undergraduate researcher in Computer Science Engineering, currently focusing on NLP, Interpretability, Social Computing, and Safety & Alignment in Large Language Models. My research aims to build pipelines that analyze biases and causal mechanisms within transformer models.

I will soon be joining the Hatzenbuehler Lab at Harvard University as a Visiting Research Fellow. Previously, I have worked with labs at Northwestern University, NTU Singapore, and the University of Hamburg.

Interests

  • Interpretability
  • NLP
  • Social Computing
  • Safety & Alignment

Education

  • B.E. in Computer Science Engineering

    Visvesvaraya Technological University (VTU), India, 2026

Experience

  • Visiting UG Research Fellow

    Harvard University (Hatzenbuehler Lab) | Mar 2026 – Jul 2026

  • Research Assistant Intern

    Northwestern University (Kellogg School) | Jan 2026 – Jun 2026

  • Research Intern

    NTU Singapore | Jan 2026 – Feb 2026

  • Research Intern (Organizer SemEval '26)

    University of Hamburg (HCDS Lab) | Aug 2025 – Present

  • Research Intern

    NIT Agartala | Jan 2025 – Dec 2025

🗞️ News

  • Feb 2026: Co-authored Agents of Chaos, a new exploratory red-teaming study on autonomous language models!
  • Feb 2026: Accepted as a Visiting Undergraduate Research Fellow at Harvard University! 🎓
  • Jan 2026: Started Research Internship at Northwestern University (Remote).
  • Aug 2025: Organizing member for SemEval 2026 Task 9: Detecting Multilingual Polarization. 🌍
  • Jan 2025: Ranked 7th (Russian) and 9th (Hindi) in SemEval-2025 Task 11 Emotion Detection.

Publications

Agents of Chaos

Natalie Shapira, ..., P Sam Sahil, ..., David Bau
Preprint (arXiv:2602.20021), 2026

POLAR: A Benchmark for Multilingual, Multicultural, and Multi-Event Online Polarization

Usman Naseem, ..., P Sam Sahil, ..., Seid Muhie Yimam
Preprint (arXiv:2505.20624), 2026

Synergizing Contextual Semantics and Moral Knowledge Graphs

P Sam Sahil, Anupam Jamatia, Kunal Chakma
Under Review

Selected Projects

Automated Circuit Analysis

Designed a fully automated pipeline for discovering, interpreting, and validating task-specific causal circuits in transformer models.

Llama-3.2 Vision Radiology

Fine-tuned Llama-3.2-11B-Vision-Instruct via LoRA for efficient adaptation to radiology image captioning using Unsloth.

AI4Democracy Platform

Monitoring platform analyzing antidemocratic narratives across 500+ daily sources using RoBERTa and Llama 2 pipelines.