mentored by Dr. Debanga Raj Neog

developed 3 pipelines for conducting mock interviews: behavioral round, puzzle-based round, and resume-based round.

integrating TTS & STT models through SarvamAI.

analyzed white papers and came up with 3 ways to reduce hallucinations for text-to-sql task

analyzed performance of 2 code LLM models under 2B parameters

mentored by Dr. Amitava Das & Aman Chadha

collaborating with a research group to experiment on mutual LLM behaviors in a multi-agent contrastive setup.

mentored by Hitesh Patel & Amit Agarwal

collaborated with researchers to benchmark LLM responses to multilingual swear word inputs.

prepared two datasets of multilingual swear words and evaluated responses from 13 SOTA open-source LLMs.

analyzed patterns and insights from the responses based on language, culture, model age, and multilinguality of LLMs.

work accepted @NAACL Industry Track '25

LINK

mentored by Dr. Kripabandhu Ghosh

experimented on the reproducibility of LLM responses in a high-stakes domain - healthcare

fetched predictions from 3 LLMs using a task prompt: Gemini 1.0 Pro, Llama 3, and Mixtral-8x7B.

extracted ReQuest algorithms from the 3 LLMs by a ReQuest prompt for the above task.

finally, retrieved LLM responses using the ReQuest algorithm with the help of a Robustness Check prompt.

LINK