Kapil Mulay

A DCGAN trained on 10,000+ brain-MRI images to synthesize tumor-positive scans, with an interactive explorer for the generator's latent space using linear (LERP) and spherical (SLERP) interpolation.

• Convolutional DCGAN generator trained on 10,000+ brain MRI images
• LERP/SLERP latent-space interpolation with an interactive explorer UI
• FastAPI + Next.js app deployed to Hugging Face Spaces and Vercel, CI/CD via GitHub Actions

A cloud-native IoT platform for water-quality monitoring — from sensor to dashboard. Electrochemical sensor nodes located at the water source publish live readings over MQTT to the nearest server, eliminating the need for manual sampling.

• Data pipeline: ESP32 sensor node → MQTT (HiveMQ Cloud) → ML Model → MongoDB Atlas via Prisma
• ML model for water-potability prediction from real-time sensor readings and a Next.js frontend for visualization.
• Automated deployment to Hugging Face Spaces through GitHub Actions CI/CD

A production-ready e-commerce platform covering the whole purchase loop — user authentication, email verification, catalog browsing, inventory management, order processing workflows, and payment gateways. This full-stack web app also supports store-preference selection by the customer.

• Relational MySQL schema for Users, Products, Orders, and Transactions
• Razorpay checkout with automated confirmation and notification emails
• Auth, email verification and store-preference selection before checkout

An autonomous micromouse that maps and solves a maze in real time with Flood-Fill, rerouting live as it discovers walls. A multi-run strategy — explore, return, optimize — picks the fastest final path.

• Flood-Fill navigation in Python and C++ with real-time rerouting, on ESP32 + ROS2
• Multi-run traversal: loop removal, diagonal path compression, regression-based feasibility scoring
• Represented IIT Ropar in the maze-solver challenge at Technoxian 2025, Noida

A collaborative Web3 project, built with IIT Ropar's Software Community, that mints verified night-sky observations as NFTs — a computer-vision check confirms a capture is genuine before it's minted on-chain.

• EfficientNet-based CV pipeline verifies a capture is a genuine night-sky observation before minting
• ERC-721 smart contracts (OpenZeppelin, Hardhat) handle minting, metadata, and ownership on Ethereum
• Next.js + Ethers.js frontend for a seamless mint-and-manage experience

A multi-agent system that simulates business negotiations and courtroom cases from user-defined system prompts, with each agent's turns streamed live.

• Orchestrated multiple LLM agents (via Groq) to role-play structured negotiation and courtroom scenarios
• Streamed agent-to-agent exchanges to the client live over Server-Sent Events (SSE)
• Persisted scenarios and transcripts with Prisma; built the interface in Next.js