Northeastern University

Indian Institute of Technology Bombay

HiDISC: Hyperbolic Domain Generalization for Generalized Category Discovery

OSLoPrompt: Bridging Low-Supervision Challenges and Open-Set Domain Generalization in CLIP

SpLIP: Elevating All Zero-Shot Sketch-Based Image Retrieval Through Multimodal Prompt Learning

Assistant Researcher — MBZUAI, Abu Dhabi , UAE

Conducted research under Dr.Muhammad Haris Khan on Multi-Label Image Classification using CLIP in a domain-generalized setting, leveraging label graphs and dynamic prompt learning to bridge domain shift , targeting a first-author publication at a top-tier AI conference or journal

Undergraduate Researcher — MLVC Lab, IIT Bombay

Worked under Dr. Biplab Banerjee on vision-language models, Domain Generalization,Generalized category discovery and image retrieval projects, resulting in publications at NeurIPS 2025, ECCV 2024 and CVPR 2025.

Student Researcher — Medal Lab, IIT Bombay

Worked under Dr. Amit Dethi on enhancing the WaveMix model, a resource efficient and performance surpassing state-of-the-art CNNs, ViTs, and token mixers for multiple computer vision tasks by setting new benchmarks for semantic segmentation on the Cityscapes dataset. Improved the model’s mIoU from 0.43 to 0.56 on ADE20K dataset for semantic segmentation by using fine-tuning methods.

Research & Development Intern — Mercedes Benz R&D India

• Optimized NLP API performance, reducing latency by 80% (from 1 minute to 10 seconds) and cutting costs by over 50% by replacing OpenAI’s GPT-3.5 with Anthropic’s Claude-3 Haiku after comprehensive LLM compatibility research
• Built and deployed CI/CD pipelines for an AI-as-a-Service (AIaaS) voice-activated scheduling assistant using GoLang, integrating Anthropic’s Claude-3 Haiku for NLP and Microsoft Graph API for dynamic data retrieval, enabling seamless two-way user interaction.

Software Engineering Intern — Shoonatyvam Foundation , DAWA.AI

• Spearheaded a team of 5 to build the startup’s various B2C products,websites,user-friendly dashboards using MERN stack and EJS
• Revamped and hosted the enterprise’s website on AWS cloud and reduced the computing costs by 10%, utilising tools like Amazon EC2 virtual cloud services, Amazon RDS for databasing, within 4 hours in collaboration with the Product and Tech teams

Data Analyst Intern — Acmegrade

Implemented an ML model and achieved an accuracy score of 88% by analyzing a Kaggle dataset of health data to determine whether the patients are suffering from Parkinson’s disease, using a support vector machine, while experimenting with other ML algorithms

Image classfication using DinoV3 embeddings | Foundations of AI project

Designing a projection head for multi-class image classification on the PACS, CUB and CIFAR-10 dataset for Generalization and utilizing the DinoV3 embeddings as features. Achieved accuracy of 99% on the PACS dataset

4 tank system simulations and fault detection | Supervised Learning Project, IIT Bombay

• Simulated an interconnected 4 tank system using MATLAB and further analysing system dynamics after introduction of time delays for various signals like PRBS
• Utilized R for data processing and creating DiGraphs for better visualization and identification of delays

Image Classification and Artifact Identification | Inter IIT Tech, Adobe

• Led a team of 10 to develop a robust Computer Vision model resilient to adversarial attacks for AI-based image identification and classification
• Researched, experimented with, and fine-tuned state-of-the-art Vision Question Answering (VQA) Vision-Language Models (VLMs) for item identification and explanation generation

Predictive capability in the sustainability data tracker of UPL | Supervised Learning Project , IIT Bombay

Designed a Machine Learning model for predicting the carbon footprint of UPL unit 7 and built an interactive Streamlit UI for gaining valuable insights for achieving Net Zero in their carbon emissions via their Data Tracker, using Python and Data Analysis

Machine Learning for Remote Sensing (GNR638) - Teaching Assistant

• Assisted in grading midterm exams and conducted tutorial sessions on RNNs, LSTMs, and multi-label classification
• Provided academic support to students by clarifying deep learning concepts and guiding them through hands-on experience in the development of machine learning models

English Language Improvement Training (ELIT) - Teaching Assistant

• Collaborated with a 14-member team to enhance the English proficiency of students at the institute
• Curated English learning resources focused on soft skills, grammar, and writing, coaching over 40 students
• Conducted weekly one-on-one sessions, providing personalized assistance to a tutee to improve English proficiency

Developer's Community (Devcom) - Core Developer

• Part of a 15-member developers crew selected via vigorous training and interviews out of 150 applicants
• Maintained critical digital infrastructure for the institute and increasing user base of InstiApp(10k+ downloads) by 15% in just 2 months by ideating product roadmaps and ongoing projects like MessI, RVP and Resobin
• Collaborated with cross-functional teams to design and implement new features, improving user engagement and satisfaction