Projects
Research Projects
Multimodal Learning for Astronomical Classification
- Designed multimodal algorithms based on CLIP and Swin Transformer architectures to classify faint & compact galaxies, stars, and quasars.
- Integrated five-channel image data with photometric features to improve feature representation and improve classification robustness.
Adversarial Machine Learning for Autonomous Vehicles
- Explored and tested over 10 unique physical adversarial attack strategies on image classification models used in autonomous vehicles.
- Developed targeted defense mechanisms, improving model robustness in simulated autonomous environments.
3D Multi-Object Tracking
- Built a transformer-based joint detection and tracking model for autonomous navigation.
- Integrated 3D LiDAR and camera data using clustering and target proposal techniques to improve multi-sensor fusion and accuracy.
Video Compression
- Developed a novel radial transform inspired by field dynamics, optimized for natural video compression.
- Designed a custom codec achieving a 1.53 dB PSNR improvement over standard methods while maintaining compression entropy.
Cross-Domain Continual Learning
- Developed a framework for federated learning that improved knowledge transfer and model adaptability across diverse domains.
Secure V2X Message Encryption
- Developed a novel V2X protocol using Key Policy Attribute-Based Encryption (KP-ABE) for secure, encrypted message transmission between vehicles.
Course Projects
Personalized Explainable Recommendations
Explainability in ML, AI5120
- Developed a transformer-based recommendation system combining Personalized Prompt Learning with Knowledge Graph Attention Networks (KGAT) for improved explainability.
- Enabled natural language recommendations by integrating user-item interactions, providing both accuracy and transparency in personalized suggestions.
News-Image Captioning
Natural Language Processing, CS5803
- Developed a unique news-image captioning model that leverages CLIP’s image and text encoders.
- Improved caption accuracy by incorporating named entities and knowledge prompts, producing context-rich captions using a multimodal large language model.
Outlier Detection
Applications of Markov Chains in Computing, CS5150
- Developed novel method leveraging Absorbing Markov Chains for saliency-based outlier detection.
Exploring Neural Network Robustness through Benford’s Law
Data Science Analysis, EP4130
- Investigated the relationship between Neural Network parameter distributions and generalization using Benford’s Law (BL) as a naturalness measure.
- Introduced the Model Enthalpy (MLH) metric to assess closeness to BL, showing its strong correlation with Validation Accuracy and robustness properties.
- Demonstrated MLH’s potential as an alternative to Validation Accuracy for Early Stopping, achieving better test accuracy without needing a validation set.
Automated Postal Record-Keeping
Software Engineering, CS4443
- Developed a text recognition system that replaced manual entry, reducing daily bookkeeping time by 85%.
Tangent - A vector graphics programming language
Compilers, CS3423
- A statically typed, object-oriented language built to create complex vector graphics like Bézier curves and polygons directly on SVGs.
- Designed from scratch using Flex and Bison
HRS Attack Simulation
Network Security, CS6903
Simulated HTTP Request Smuggling attacks with novel detection and mitigation techniques.
Developer Projects
Trumio’s AI-Driven Testing & Mentorship Platform
- Created an alumni mentoring and peer-testing platform featuring AI-powered tester matchmaking, explainable recommendations, and AI-generated test suites.
- Developed a comprehensive business plan for the same.
Sentiment Analysis
- Built an alumni mentoring and peer-testing platform with AI-powered tester matchmaking.
- Integrated explainable recommendations for tester assignments and AI-generated test suites.
GrowSimplee’s Route Planning for Optimized On-Time Delivery
- Delivery Route Optimization Task based on Package characteristics and delivery points.
- Implemented a DNN in conjunction with a weight sensor to determine the dimensions of object from image
Cryptex
- Developed and optimized the Cryptex (Online Treasure Hunt) platform, integrating advanced admin features for smooth moderation and supporting a global user base.
Campus Portal
- Developed a comprehensive portal for the college community, featuring cab-sharing, marketplace, lost & found, and other resources for campus connectivity.
TEDx IIT Hyderabad
- Website Development for TEDx chapter of IIT Hyderabad for the year 2021 - 2022.
- Developed using Jekyll.
EBSB IIT Hyderbad
- Website Development for EBSB for the year 2023-2024.