Vethavikashini C R

I am an Applied Scientist II at Amazon's US Prime and Marketing Technology (UPMT) organization, where I work on large-scale personalization and recommendation systems. I completed my MS at Columbia University in 2024, advised by Prof. Micah Goldblum, Prof. Zhou Yu, and Prof. Siddhartha Dalal. During my academic journey, I have also had the privilege of being mentored by Prof. Erik Cambria and Prof. Ashish Anand.

I graduated from IIT Guwahati in 2023 with a Bachelor's in Engineering Physics. Previously, I interned at Adobe Media Data Science Labs, where I worked on user behavior modeling for large-scale media experiences.

Amazon Email  /  Personal Email  /  Scholar  /  Twitter  /  Github  /  LinkedIn

Research

My primary research interests span four key areas:

• AI Safety and Privacy: I am passionate about developing robust systems to ensure the safe deployment of AI technologies. My work focuses on building privacy-preserving frameworks for large language models, designing secure agentic systems, and addressing vulnerabilities like adversarial attacks and inference-time privacy leakage.
• Causal NLP: I aim to enhance the interpretability and fairness of language models by leveraging causal inference techniques. My research includes uncovering and mitigating biases in generative models and exploring causality-driven methods to improve fairness and decision-making in NLP.
• Large-Scale Personalization and Recommendation Systems: At Amazon, I work on building and evaluating machine learning systems for personalized product discovery at scale, serving hundreds of millions of customers globally.
• NLP for Social Good: I am dedicated to using NLP to address real-world challenges, including creating inclusive AI tools for low-resource languages, evaluating AI-powered educational systems, and tackling issues related to climate change and community well-being.

News and Highlights

• [April 2026] Best Poster Runner-Up Award at AIX Summit East 2026 for Bringing Pedagogy into Focus: Evaluating Virtual Teaching Assistants' Question-Answering in Asynchronous Learning Environments
• [November 2025] Bringing Pedagogy into Focus: Evaluating Virtual Teaching Assistants' Question-Answering in Asynchronous Learning Environments accepted at EMNLP 2025 Findings
• [April 2025] PAPILLON: PrivAcy Preservation from Internet-based and Local Language MOdel ENsembles accepted at NAACL 2025, Long Papers
• [April 2025] Causally Testing Gender Bias in LLMs: A Case Study on Occupational Bias accepted at NAACL 2025 Findings
• [February 2025] Commercial LLM Agents Are Already Vulnerable to Simple Yet Dangerous Attacks released on arXiv
• [December 2024] Awarded Best Paper for Causally Testing Gender Bias in LLMs: A Case Study on Occupational Bias at Causality and Large Models - NeurIPS 2024

Publications

* denotes equal contribution.