CV

Research Assistant

Artificial Scientific Intelligence Lab, National University of Singapore

• Develop research on reward-proportional reinforcement learning and outcome-level mode collapse, resulting in a first-author ICML 2026 publication and an AAAI 2027 submission.
• Designed density-aware reward scaling for continuous outcome spaces and integrated it with GRPO for multimodal navigation and language-model-based text-to-image prompt adaptation.
• Developed methods for LLM post-training and alignment with emphasis on reward design, diverse generation, and reasoning-oriented evaluation.
• Developed an RL agent that edits discrete VQ-VAE latent codes using classifier-confidence and discriminator-realism feedback.

Machine Learning Engineer

VAO Labs

• Developed LayoutLM and OCR document-understanding pipelines for structured information extraction.
• Fine-tuned T5-family models for document question answering and answer retrieval.
• Built Llama-based applications for querying large enterprise datasets with natural language.

Research Assistant

Computational Neuroscience Lab, IIT Madras

• Designed a brain-inspired visual attention model that learned saccadic scanning policies with Q-learning and recurrent neural networks.
• Developed computer-vision and reinforcement-learning systems for neurological healthcare applications.
• Invented A System for Monitoring Parkinson’s Disease, granted as an Indian patent to IIT Madras.

Software Developer

IIFL

• Developed full-stack software using ASP.NET, SQL Server, Angular, JavaScript, and jQuery.
• Integrated, tested, and documented changes using TFS and Azure in an agile workflow.

PhD

National University of Singapore

Machine Learning

• Thesis: Diversity and Creativity in Generative AI
• Supervisor: Dr. Dianbo Liu

Dual Degree (Bachelor's and Master's)

Indian Institute of Technology Madras

Biotechnology

Expected Return Causes Outcome-Level Mode Collapse in Reinforcement Learning and How to Fix It with Inverse Probability Scaling

International Conference on Machine Learning (ICML)

Explains outcome-level mode collapse induced by expected-return objectives and introduces inverse-probability scaling to recover diverse high-reward outcomes.

Density-Aware Reward Scaling: A Reinforcement Learning Framework for Reward-Proportional Sampling

Submitted to AAAI 2027

Introduces DARS, a critic-free method that scales rewards by local outcome density while retaining the standard policy-gradient objective.

How Does My Model Fail? Automatic Identification and Interpretation of Physical Plausibility Failure Modes with Matryoshka Transcoders

Spotlight, CVPR XAI4CV Workshop

Introduces hierarchical sparse features for coarse-to-fine interpretation of physical-plausibility failures across generative models.

Brain-Inspired Attention Model for Object Counting

International Conference on Neural Information Processing (ICONIP)

Developed a Q-learning-based sequential attention model with recurrent visual glimpses, achieving 92.1% object-counting accuracy.

Editing Discrete Latent Variables with a Reinforcement Learning Agent

An RL framework that edits discrete VQ-VAE codes toward a target class while minimizing codebook changes.

• Combines classifier-confidence and discriminator-realism rewards for controlled latent-space editing.
• Produces multi-step trajectories that reveal interpretable, task-dependent latent navigation.