About
I am a researcher at IIT Hyderabad, advised by Dr. Vineeth N. Balasubramanian. My research sits at the intersection of Continual Learning, Long-Tailed Classification, and Large Language Models, with a focus on building AI systems that learn realistically — without unrealistic assumptions about data distribution or task boundaries.
I recently completed an internship as an Applied Scientist at Amazon India, where I developed new capabilities for agentic AI systems and built evaluation frameworks that reduced testing effort by 7×. I am a Reliance Foundation Fellow (Top 100 nationwide).
Education
Indian Institute of Technology, Hyderabad
Master of Technology in Computer Science & Engineering (By Research)
CGPA: 9.76 / 10
Jan '23 – Dec '25
Publications
Class Incremental Learning Free of Unrealistic Assumptions
Structure Thinking for Enhanced Reasoning in LLMs
Transitioning Heads Conundrum: The Hidden Bottleneck in Long-Tailed Class-Incremental Learning
A Deeper Look at the Hessian Eigen Spectrum of Deep Neural Networks and its Applications to Regularization
Research Experience
Applied Scientist Intern
Amazon · India
Mar '25 – Aug '25
- Developed a new capability for Amazon's internal agentic assistant (SAPIEN), enabling insight exploration across databases and user files through automatic table identification (89% accuracy), context-aware clarification handling, and workflow initiation; demoed to leadership.
- Reduced manual testing effort of SAPIEN (7×) by building tools like Golden Dataset Generator and Gamma Testing Framework, ensuring scalable reliability across use cases.
- Secured a Top-3 finish (only intern to do so) at Amazon's internal Hackathon with Promptinator-3000, an automated prompt and dataset generation framework achieving 21× efficiency gain, earning recognition from leadership.
★ Under Review: ACL Rolling Reviews '26
Jan '23 – Dec '25
- Proposed a new realistic setup in Continual Learning that is free of assumptions and developed a novel method using adapters to tackle it, yielding a +8% performance boost.
- Developed a novel method to tackle the Transitioning Head problem in Long-Tailed Class Incremental Learning via Early Knowledge Transfer, achieving state-of-the-art results (+5% performance).
- Created a human-in-the-loop recourse framework integrating user feedback, generating personalized counterfactuals and enhancing user satisfaction. Achieved a 13× improvement in metrics over baselines.
★ Published: AAAI '26 (W), TMLR '26, TMLR '25 · Under Review: ECCV '26
Research Intern
IIT Hyderabad · India
Advisors:
Dr. Vineeth NB (IIT-H) &
Dr. Makarand Tapaswi (IIIT-H)
Jul '19 – Jan '23
- Developed TailCalibX, a feature generation technique for Long-Tailed classification that uses calibrated distributions to boost performance on imbalanced datasets, setting a new state-of-the-art.
- Created a Hessian-based regularization method that improves generalization by leveraging the similarity between layerwise and overall Hessians, enhancing regularizer efficiency.
★ Published: AAAI '21, ICVGIP '21
Research Projects
Integration Testing for Stochastic AI Systems
May '25 – Aug '25
Advisor: Anubhav Shrimal (Amazon)
Developed an LLM-based, persona-driven multi-turn evaluation framework for stochastic AI systems, simulating real user behaviors (waiting, task switching, file uploads) and enabled headless CI/CD integration to produce automated pass/fail signals, improving reliability of validation.
Adversarially Coupled Prompt & Dataset Generator
Mar '25 – May '25
Advisor: Anish Nediyanchath (Amazon)
Developed an adversarial co-evolution framework coupling a prompt generator and dataset generator with access to intermediate reasoning steps, achieving a 21× efficiency gain by producing robust initial prompts and diverse reasoning-annotated datasets, reducing manual effort and accelerating UAT cycles.
AWARE: Adaptive Wear-Levelling and Attack Re-mapping Engine
Jun '24 – Dec '24
Advisor: Dr. Shirshendu Das (IIT-H)
Proposed a novel framework that enhances NVM durability and security in LLCs by combining adaptive wear-leveling and attack mitigation through intelligent remapping and workload-aware strategies.
Large Language Models Related Hardware Optimizations
Jan '24 – Apr '24
Advisors: Dr. Rajesh Kedia & Dr. Shirshendu Das (IIT-H)
Analyzed hardware and software optimizations to improve Large Language Models, identifying gaps like fragmented benchmarking and advocating for unified solutions to boost efficiency.
Achievements
'23 – '25
Reliance Foundation Postgraduate Scholarship
Awarded to Top 100 Students Nationwide. Scholarship Value: ₹6 Lakhs.
'26
AAAI Travel Grant Awardee
Competitive grant supporting travel and accommodation to attend AAAI 2026.
'25
Amazon Internal Hackathon — Top-3
Only Intern Finalist, competing against well-seasoned Applied Scientists & SDEs.
'24
Amazon Machine Learning Summer School
Selected participant in the Amazon ML Summer School program.
Service & Volunteering
Research Service
Reviewer: AAAI '26, ECCV '22
Sub-Reviewer: CVPR '23, ICLR '21, IJCAI '20, WACV '23, SDM '21
Student Volunteer: AAAI '26, ACML '22, ICML '20
Teaching Assistantships
AI and Emerging Technologies — IIT Hyderabad
('26, '25, '24, '23, '22)
Deep Learning for Computer Vision — NPTEL
('24, '20)
Effective Teaching of Machine Learning — CSEDU IIIT Delhi
('22, '21)
Advanced Topics in Machine Learning (AI 2100 / CS 6360) — IIT Hyderabad
('21)