Smart Vehicle Intelligence Assistant with MCP

• Built comprehensive vehicle management platform using React, TypeScript, and OpenAI GPT-4o for real-time health monitoring, predictive maintenance, and conversational AI assistance
• Integrated DIMO Network APIs with blockchain-based authentication, implemented Model Context Protocol (MCP) for secure vehicle data access and verifiable credentials
• Developed OCR-powered document processing with AI analysis for maintenance receipts, enabling automatic service scheduling and cost optimization
• Deployed production-ready solution on Vercel/Railway with Supabase backend, featuring real-time telemetry, multi-vehicle dashboard, and responsive design serving live users

DIMO Road Racer, Web3 Gaming Platform

• Built DIMO Road Racer, the first-ever Mini App on Coinbase's BASE network - Developed a web-based racing game using Next.js, TypeScript, and HTML5 Canvas with blockchain wallet integration, daily challenges, and real-time leaderboards featuring 150K+ connected wallets
• Engineered complete Web3 gaming ecosystem - Implemented upgradeable Solidity smart contracts (UUPS proxy) for automated $DIMO token rewards, integrated Coinbase OnchainKit for seamless wallet connections, and built Supabase-powered leaderboards with daily streak systems and gift box rewards
• Scaled to 150K+ connected wallets on BASE network - Launched as the first Mini App on Coinbase's BASE network, powering daily competitive gameplay with real $DIMO token rewards (1000/500/250 tokens for top 3 players), automated smart contract distributions, and 87+ production deployments serving active gaming community

AI Diffusion Model for Non-Fungible Token (NFT) Validation

• Developed an AI-powered NFT generation pipeline using Stable Diffusion and CLIP for prompt-conditioned image synthesis and semantic relevance scoring
• Engineered a blockchain-ready validation system with perceptual hashing to enforce NFT uniqueness, enhancing trust and preventing duplicate minting
• Deployed end-to-end system with a Flask backend on Render and React frontend on Vercel; integrated Replicate API for scalable inference and real-time NFT generation
• Achieved 93% average CLIP relevance score and 98% duplication prevention across 1000+ test prompts, with 60% reduction in image generation latency through optimized API integration

Customer Churn Prediction System on Google Cloud Platform

• Built and deployed a high-performance churn prediction model using a Stacking Ensemble framework that combined Decision Tree, Random Forest, and Gradient Boosting models, achieving 90% accuracy and 0.92 AUC-ROC for real-time predictions
• Implemented comprehensive data preprocessing techniques, including SMOTE for class balancing, feature encoding, normalization, and outlier handling, ensuring model robustness, and improved predictive accuracy for banking sector applications

Stock Movement Using State-of-the-art Transformer Model

International Conference on Advances in Computer Engineering, Communication Systems and Business Development (ICACECSBD-2023), Article number 18, ISBN: 978-93-92105-64-7

• Built and compared the transformer model using Python for stock price prediction with two other models namely ARIMA and LSTM
• Evaluated the three models through various parameters after training and testing, the validated score of Mean Square Error of 0.0004091 (least for transformer model) was obtained for the given Amazon Stocks Time series dataset

Convolution Neural Network for Glaucoma Detection

Institution of Electronics and Telecommunication Engineers (IETE-SF 2023) Spark Journal, Article number 20, ISBN: 978-93-5777-300-3

• Developed CNN-based DenseNet-121 and ResNet-50 models for detecting glaucoma from extracted images
• Achieved an average F-measure of 85% and accuracy of 74% by DenseNet-121 model, whereas the ResNet 50 model and achieved an average F- measure of 83% and accuracy of 72% in distinguishing glaucoma positive from healthy eye

2021's Indian Initial Public Offering analysis using Python

Institution of Electronics and Telecommunication Engineers (IETE-SF 2022) IGNITE Newsletter, Article number XV

• Created a detailed data analytical report of Indian IPO which proved Indian Stock market as one of the best performing markets in Asia for the financial year 2021 using Python

Wideband and Compact Rectangular Shape Microstrip Antennas Employing Rectangular Slots Cut Bow-Tie Shape Ground Plane

Progress in Electromagnetics Research B (PIER B) Journal 2023, Vol. 100, page 155-172, ISSN: 1937-6472

• Patch employed three rectangular slots cut Bow-tie shape ground plane, yield optimum results. Applied the rectangular patch, bandwidth increased by 20% and resonance frequency, substrate thickness, and patch area reduction by 32%, 0.034λg, and 61.12%, are respectively achieved
• Proposed antennas exhibit radiation pattern in the broadside with a gain of more than 5 dBi

Compact Designs of Circular Microstrip Antennas Employing Modified Ground Plane for Wideband Response

Elsevier's AEU - International Journal of Electronics and Communications, Volume 176, Article 155130, March 2024, ISSN: 1618-0399

• Ground plane slots reduces patch fundamental mode frequency and Bow-tie shape of the same yields wider bandwidth on thinner substrate, thus achieving wider bandwidth with a smaller antenna volume
• Modified ground plane antenna yields 25% bandwidth increment, 0.03 λg reduction in substrate thickness, 35.12% reduction in the resonance frequency, 66% reduction in the patch area with a broadside peak gain of 5.5

Accident Alert System Using GPS And GSM Technology

Institution of Electronics and Telecommunication Engineers (IETE-SF 2022) IGNITE Newsletter, Article number XV

• Combined a GSM and GPS interface with an Arduino Uno board as a platform, can make vehicle tracking more efficient and convenient as it can work in any weather and offer real-time position along with the co-ordinates
• The geographic location of the coordinates are automatically displayed on register phone number by just one manual push on button