Hello! I'm Bagyasree.

My overall aim is to conduct research to improve language models’ social and moral intelligence, and thus make them better at and safer for complex interactive tasks like teaching and psychological interventions. To this end, I’m particularly interested in:

1. Helping LMs reason about implicit human behaviour and language, and aligning them with pluralistic human values.
2. Defining realistic and model-friendly specifications, benchmarks, and measures for these values.
3. Integrating ideas from linguistics, cognitive science, and philosophy, as well as findings from social simulations, with traditional NLP pre- and post-training methods to achieve the above.
4. Exploring how these ideas can be used to make models generalizable across social domains.

I am currently interning with the ML R&D team at Vectara, where I’m optimizing the efficiency of agentic evaluation suites, tying user satisfaction signals to evaluation outcomes, and making technically rigorous evaluations user-friendly and digestible. I graduated with a Master’s in Computer Science from CU Boulder in December 2025, where I was part of the BLAST lab led by Dr. Maria Pacheco, working on computational social science. I’m also working with Dr. Alvaro Velasquez to explore sycophancy in large language models. I got my undergraduate degree from PES University, Bengaluru. I’ve worked on several interesting AI projects since I started studying CS — I’m looking for PhD positions, to start in 2027!

I love books and stories; I think fiction is our richest source of information about the human condition, and one of the best ways to develop empathy for the world. Plus, they’re fun! I’m currently obsessed with all things Cosmere. Sometimes I post reviews about the books I’ve read on Goodreads. I’m also a bit of a crossword nerd, especially cryptics.