*
here are some of the projects i've built. i focus on machine learning, data science,web development and python-based tooling.
creator (2026)
A fully self-hosted music streaming ecosystem combining a Navidrome backend, a Feishin web player, and a custom 512-dimensional Vector AI recommendation engine. It features seamless integration with Azure Blob Storage for infinite scaling, a custom dual-bot Telegram architecture using Python and Telethon for automated Spotify playlist importing, and a FastAPI middleware for Subsonic-compatible AI endpoint interception and dynamic AutoDJ curation.
achievements
- Built a 512-dimensional Vector AI recommendation engine using Cosine Similarity to mathematically cluster tracks by audio texture, mood, and semantic metadata
- Optimized Python machine learning inference with high-speed NumPy matrix multiplication to compute similarities across 1,300+ tracks instantly
- Designed a custom dual-bot Telegram architecture using Telethon to automatically intercept and download Spotify playlists directly to server SSDs
- Integrated Azure Blob Storage via blobfuse2 to bypass VM disk constraints, enabling infinite and cheap library scaling
- Reverse-engineered the Feishin AutoDJ API via HAR analysis to intercept network requests using custom FastAPI middleware
- Secured the entire ecosystem with Cloudflare Tunnels and Docker for safe, port-free remote access and SSL termination
technologies
creator (2026)
ai-powered civic complaint platform for mumbai citizens — report infrastructure issues via photo, get an auto-generated formal complaint routed to the correct bmc department, and track resolution status in real time
achievements
- built a multi-model ai vision pipeline using gemini 2.5 flash, gemini 2.5 flash-lite, and gemini 3.1 flash-lite in a waterfall fallback chain — automatically falls back to groq llama-4 scout vision if all gemini daily quotas are exhausted
- implemented groq llama-3.3-70b for generating formal bmc complaint letters with correct ward and department addressing
- designed a flask monolith backend with jwt authentication via httponly cookies, bcrypt password hashing, nosql injection prevention, ip-aware rate limiting, and input sanitization across all endpoints
- built next.js 16 app router frontend with edge middleware jwt verification using jose — protects routes server-side before page renders, zero client-side token exposure
- architected cross-domain cookie strategy between vercel frontend and render backend — frontend sets same-domain cookies from jwt in json response for middleware routing while render httponly cookies authenticate actual api calls
- built a separate admin dashboard with role-based jwt claims, admin-only complaint management, status updates reflected live on user dashboards, and aggregate analytics across all wards
- added a public community feed showing all city-wide reports with cloudinary images, severity and status badges, typed location text, and one-click google maps navigation via gps coordinates
- implemented a keep-alive strategy using uptimerobot and cron-job.org pinging a lightweight health endpoint every 5 minutes to prevent render free tier sleep — 540+ daily ai analyses available across model quota buckets before groq fallback triggers
- deployed frontend on vercel and backend on render with full environment variable separation, cloudinary image storage, and mongodb atlas with compound indexes for query performance
technologies
creator (2026)
Sage is a two-tier AI-powered productivity web app with a Next.js frontend and a Flask backend. It features Google OAuth login, a daily AI briefing powered by Gemini, a natural language command interface ("Ask Sage") that supports text and image/PDF input via Groq LLM, Google Calendar sync, task and expense CRUD, and a streak tracking system. Data is persisted in MongoDB across users, tasks, expenses, and streaks collections.
achievements
- Built a full NLP pipeline using Groq and Gemini APIs to parse natural language commands into structured task/expense actions
- Integrated Google Calendar and Gmail APIs for real-time calendar sync and email action extraction
- Implemented JWT-based auth with Google OAuth 2.0 and protected route middleware on both frontend and backend
- Designed a multi-collection MongoDB schema (users, tasks, expenses, streaks) with aggregation pipelines for finance summaries
- Built a daily AI briefing endpoint that synthesizes tasks, calendar events, bills, and email data into a structured summary
- Developed 20+ REST API endpoints across auth, tasks, finance, NLP, briefing, and streak domains