cv

Research Assistant

GAMMA Lab, University of Maryland, College Park

Curated a dataset of 1,000+ instances of non-speech sounds and music paired with multiple choice questions requiring audio-reasoning skills for DCASE2025. Actively working on music intelligibility and understanding.

• Dataset creation for non-speech sounds and music
• Audio reasoning QA for DCASE2025
• Music intelligibility research

Master’s Thesis Student

Multimodal Perception Lab, IIIT-Bangalore

Proposed a novel video-audio embedding scheme to capture implicit expressions and dialogue intent in acting auditions. Improved state-of-the-art performance by 12%.

• Weakly supervised multimodal learning
• Sequence analyzer for audition evaluation

Machine Learning Research Intern

Adobe Research

Developed algorithms for textual hierarchy estimation in design documents. Built GPT-based pipelines for generating contextual prompts to populate design templates.

• Algorithm for document hierarchy estimation
• LLM-based design prompt generation
• 80% human acceptance in evaluations

Machine Learning Research Intern

E-Health Research Center, IIIT-Bangalore

Built an attention-based video model for identifying self-stimulatory actions from raw videos, with 30% lower inference latency and 81% accuracy. Authored a paper published at IEEE Healthcom 2023.

• Attention-based video modeling
• Self-stimulatory behavior detection
• IEEE Healthcom 2023 publication

M.S.

University of Maryland, College Park

Computer Science

B.Tech + M.Tech

International Institute of Information Technology, Bangalore

Computer Science (Specialization: AI & ML)

MMAU-Pro: A Comprehensive Benchmark for Holistic Evaluation of Audio General Intelligence

AAAI

Comprehensive benchmark for evaluating general audio intelligence across multiple tasks. Submitted to AAAI 2026 Main Track.

Introducing SSBD+ Dataset with a Convolutional Pipeline for Detecting Self-Stimulatory Behaviours

IEEE Healthcom

Proposed a convolutional pipeline for detecting self-stimulatory behaviours from videos, introducing the SSBD+ dataset.

GPT-3 Implementation

Implemented GPT-3 architecture from scratch using PyTorch and trained on the Tiny Shakespeare dataset.

• Self-attention trick
• Pretraining experiments

Van Gogh Me!

Generated Van Gogh-style celebrity portraits using CycleGAN trained on Van Gogh paintings and CelebA dataset.

• CycleGAN
• Image-to-image translation