Selected Projects
AI pipelines, autonomous agents, data systems, and production builds.
Click Technical Details to expand the engineering breakdown for each project.
Agentic RAG Pipeline
LangChain Agents, ChromaDB & Llama Fine-Tuning
A fully autonomous AI system that ingests documents, reasons across them in multi-step chains, and generates context-aware responses — running entirely on local hardware at zero cloud cost.
▸ 500+ post training dataset · Zero cloud cost · Full local GPU stack · MLflow experiment tracking
Geospatial Predictive Maintenance
Anomaly Detection on Commercial Fleet Telemetry
An ML system that predicts mechanical failures in commercial truck fleets before they happen — using real sensor data and geospatial mapping to show where failures cluster across regions.
▸ F1-score: 0.82 · SMOTE class balancing · Interactive geospatial risk maps · Reduced manual fleet monitoring
Breast Cancer Detection
Medical Imaging CNN Pipeline
A deep learning classifier that detects cancerous tissue from medical images — trained end-to-end with a full evaluation dashboard for model benchmarking and inference testing.
▸ 96.5% validation accuracy · CUDA-accelerated training · Streamlit inference UI
AI Color Palette Generator
Computer Vision Pipeline + Gemini Vision API
A full-stack AI web app that takes any image upload and generates a branded colour palette — from dominant colour extraction to final hex codes — in a single automated pipeline. Built and deployed in under 2 weeks.
▸ End-to-end in < 2 weeks · REST API containerised with Docker · Multi-step agentic tool pipeline
Field Trip Fighters — Adaptive RL Boss AI
Hybrid Q-Learning + PPO · Published Research Paper
A fighting game AI boss that learns from every match — adapting its strategy in real-time to the human player's patterns. Built as a capstone project and published as a peer-reviewed research paper.
▸ 95% win rate vs beginner AI · 72% vs advanced AI · 2,590 training episodes · Published at NMIMS MPSTME
Car Rental Database System
Normalized MySQL Schema for Shared Mobility
A fully normalized relational database for a car rental platform — handling fleet inventory, customer accounts, real-time bookings, and operational analytics. Built to support high-traffic transactional workloads.
▸ Fully normalized schema · Optimized JOIN queries · Stored procedures + triggers for data integrity
AI-Based Timetable Detection
Computer Vision + Gemini Vision API
An automation system that reads a photo of a timetable and converts it into structured, database-ready data — eliminating manual data entry through a computer vision + AI pipeline.
▸ Unstructured image → structured schema · End-to-end automation · Reliable post-processing validation
IEC NMIMS Website
Production Full-Stack Web Application
The official website for the Innovation and Entrepreneurship Cell at NMIMS — a production-grade web app built, deployed, and maintained by the technical team I led as VP and Head of Technical Department.
▸ Live at iecnmims.com · Used by 500+ students · Production deployment with CI/CD
Student Portal Mobile Application
Android App with Real-Time Firebase Backend
A mobile app for university students featuring authentication, a notice board, task management, and faculty access — backed by real-time Firebase data synchronization.
▸ Firebase real-time sync · Full auth system · Notice board + task management
VR Gun Duel Game
Unity / C# with Image Processing Backend
A real-time VR gun duel game built in Unity/C# with a backend service that processes image inputs to extract and validate structured game data — bridging computer vision with game systems.
▸ Real-time C# system · Image processing backend · Runtime performance optimization
Loan Default Prediction
ML Risk Scoring Model — Inditrade Capital
A machine learning risk model built during an internship at a financial services firm — predicting high-risk loan defaults from historical data, with outputs directly used by the collections team.
▸ 78% precision on high-risk accounts · 30% improvement in risk identification · Deployed to collections team