I like teaching machines to read the room, to understand what's really being said and reply the way a real human would. That curiosity is what pulled me from an undergraduate research lab in Pilani into Microsoft's PROSE team, and now into a Master's at Georgia Tech.
I'm currently an MS student in Computer Science at Georgia Tech and a Graduate Research Assistant in the NLP-X Lab, advised by Prof. Wei Xu. My current research sits at the intersection of LLM evaluation, multilingual NLP, and user simulation. I'm especially interested in making sure LLMs understand what a user really means beneath the words and reply in a way that feels genuinely human.
Before Georgia Tech, I spent two years as a Research Engineer on Microsoft's PROSE team, where I built synthetic-data and fine-tuning pipelines for automatic code generation. I was drawn to what happens when models run out of clean, human-labeled data and have to learn from data they (or their peers) generate themselves, as well as grounding LLMs in structured knowledge and symbolic reasoning. I was fortunate to be advised there by Dr. Jose Cambronero, Dr. Sumit Gulwani, and Dr. Aditya Kanade.
I earned my bachelor's degree in Computer Science from BITS Pilani, graduating in the top 1% of my class with the Dean's Merit Scholarship, and picked up research experience along the way at CSIR-CEERI, TU Darmstadt under Prof. Iryna Gurevych, and American Express AI, with undergraduate advisors Prof. Yashvardhan Sharma and Prof. Pratik Narang. Most recently, I spent this past summer as a Machine Learning Scientist Intern at JPMorgan Chase, building LLM agentic pipelines that adapt their shape and workflow to cost, quality, and latency constraints.
- Built an LLM agent that dynamically adapts its shape and workflow based on cost, quality, and latency constraints.
- Designed a news analytics platform that retrieves articles by intent and surfaces entity-level sentiment on financial themes.
- Built the first pragmatic-evaluation benchmark for Indic languages, revealing LLM failures that existing benchmarks miss.
- Designed an LLM-based human-learner simulator that mimics real human emotion, intent, and style via RLHF.
- Built an evaluation framework to measure LLM fidelity in simulating multi-turn human learning dynamics.
- Developed NL validation strategies to obtain high-quality synthetic data for fine-tuning LLMs in low-resource settings.
- Managed and deployed an end-to-end pipeline for fine-tuning LLMs on the NL-to-spreadsheet-formula task in Excel Copilot.
- Curated an evaluation dataset for LLM-based code auto-completion on long-context, semi-structured data.
- Trained four Transformer-based ML models for credit fraud prediction over tabular and continuous features.
- Adapted these techniques to large-scale industry data, improving performance by 3% over the base LightGBM approach.
Journey
A self-improving LLM framework that distills context into reusable implications instead of query-specific answers, matching SEAL's accuracy with 20× less training time and improving reasoning accuracy by up to 10.5 points.
Two risk-aware variants of the TD3 algorithm that let an agent explicitly trade off average performance against worst-case outcomes, outperforming standard TD3 in MuJoCo environments with rare but severe failures.
A custom compiler implementing every phase from parsing to assembly code generation, built from scratch in C.
-
DAAD-WISE ScholarshipDAAD, Germany
Selected among the top 100 students in India for a fully-funded (~$4,000) research internship in Germany.
-
NTSE ScholarshipGovt. of India
Ranked in the top 750 among 200,000+ applicants on the National Science Talent Examination.
-
ACM Student GrantACM CODS-COMAD
Received a travel grant worth Rs. 30,000 from ACM to attend the CODS-COMAD conference.
-
Teaching Assistant, Georgia Tech
TA for Machine Learning.
-
Executive Board, Grad Pride @ Georgia Tech
Helping organize community and events for LGBTQ+ graduate students on campus.
-
Teaching Assistant, BITS Pilani
Undergraduate TA for Computer Programming and Data Structures & Algorithms.
-
Student Faculty Council, BITS Pilani
Student representative in the Department of Computer Science.
-
Project Leader, Nirmaan Organization
Guided a team of volunteers teaching computer skills to underprivileged youth.
Beyond research: I'm usually reading about geopolitics and history, planning the next trip, experimenting in the kitchen, or in the middle of a friendly debate.