A collection of my work in AI optimization, low-level systems, and full-stack development.
Rail Yatri
Tech Stack
FastAPI Neo4j Three.js ML
Smart railway dispatcher developed for SIH 2025 (Top 30 out of 125+). Uses MILP and Deep Reinforcement Learning to resolve real-time scheduling conflicts for the Mumbai Railway network.
HTTP over TCP
Tech Stack
Go TCP Systems
A high-performance HTTP/1.1 server built from scratch in Go, bypassing standard libraries to implement core networking protocols directly on raw TCP streams.
Judiciary AI
Tech Stack
RAG Flask React Gemini
An AI-driven legal assistant using PDF vectorization and RAG to perform semantic search across complex Indian judicial datasets. Ranked 4th in SIH internal qualifiers.
RaAI
Tech Stack
MongoDB LangChain Docker LLMs
Awarded 2nd place at HackXplore '25. An emotional intelligence agent using RAG to analyze journals and detect cognitive distortions using multi-LLM orchestration.
EventHub
Tech Stack
PostgreSQL Django Next.js
A full-stack event management system built for the committees of MPSTME. Allows users to discover events, register, and manage complex committee activities.
Chirpy
Tech Stack
Go PostgreSQL JWT Webhooks
A high-performance Go backend for a micro-blogging platform. Features automated profanity filtering, role-based upgrades via webhooks, and administrative metrics tracking.
Clerk
Tech Stack
LangChain FastAPI Next.js NLP
A voice-driven AI agent that transcribes consultations and drafts clinical notes. Integrates PrimeKG knowledge graph for live drug interaction support.
Alzeimers detection
Tech Stack
Python EfficientNet NumPy PyTorch
Deep learning system using EfficientNet-B2 to classify Alzheimer's disease progression from MRI scans. Achieved 86% accuracy on the OASIS-2 dataset.
DevGaurd
Tech Stack
JavaScript Chrome API REST API
A browser extension that redirects traffic to LeetCode until the daily problem is completed. Leverages Chrome declarativeNetRequest and LeetCode APIs.
Ad Targeting
Tech Stack
DBSCAN Softmax Flask ML
Ad-serving platform using DBSCAN for user segmentation and Softmax for purchasing intent. Awarded 2nd place at ACE 1.0; uses Gemini for behavioral-targeted captions.