S I Harini

I am a Research Associate 2 at Adobe MDSR. My research explores how AI can help us understand and predict human behavior—and how insights about human behavior can, in turn, make AI systems more effective. I pursue this through two complementary threads: developing AI systems that learn from human behavior to understand, predict, and optimize how people respond to content, and developing LLM-based agents that can accelerate scientific discovery by uncovering novel, empirically grounded theories.
Previously, I earned my bachelor's in CS from BITS Pilani, Goa, where my undergraduate thesis with Prof. Nick Haber and Fan-Yun Sun at Stanford focused on generating end-to-end game and robotics environments from natural-language input to train reinforcement learning agents. At BITS, I was mentored by Dr. Ashwin Srinivasan and Dr. Gautam Shroff, where I explored RL and neurosymbolic techniques to solve the ARC-AGI challenge.

Google Scholar  /  GitHub  /  Resume  /  Twitter  /  Linkedin

• 2026-08 Exploring Collaboration between a language and a non-language agent, accepted to EMNLP Main Conference!
• 2026-06Our paper ZIPP: Zero-Shot Image Personalization from Personas is accepted to ECCV'2026! See you in Sweden!
• 2026-02Our paper Accelerating Social Science Research via Agentic Hypothesization and Experimentation is on arXiv!
• 2026-02 Got promoted to Research Associate 2 at Adobe MDSR!
• 2026-01 Unsupervised Memorability Modeling from Tip-of-the-Tongue Retrieval Queries , is accepted to WACV'2026!
• 2025-01 Behavior LLava and Transsuasion accepted to ICLR'2025!
• 2024-10 Long-Term Ad Memorability: Understanding & Generating Memorable Ads, is accepted to WACV'2025.
• 2024-10 Released PersuasionArena A large-scale automated leaderboard of LLM's persuasion capabilities.
• 2024-10 FactorSim, is accepted to NeurIPS'2024.
• 2024-07 Graduated from BITS Pilani with a Bachelor's degree in Computer Science.
• 2024-05 Released BehaviorLLaVA, training VLMs on human behavior improves performance on close to 50 tasks.
• 2023-05 Joined Adobe as a research intern, working on multi-modal foundation models

Coming soon...