Ideas worked out in code: machine learning models, agents, and full-stack apps
Turns Instagram reels into schema-validated recipes, itineraries, and workouts (plus PDFs) by combining audio, caption, and on-screen text as separately cited evidence. Includes adaptive extraction budgets, tiered LLM provider chains with a hard per-reel cost cap, hallucination checks backed by an eval harness, SSE progress streaming, and an iOS/Android app built with Capacitor and an Instagram share extension.
Model Context Protocol server simulating Pokemon battles, built for AI agent integrations and tool-use demonstrations.
A platform that keeps groups connected with built-in calendar, feed, task management, and shared expenses ledger.
Full-stack churn prediction app. The LSTM backend forecasts gym attendance with 85% accuracy.
Distributed pipeline for estimating muscle force from high-density electromyography signals with 3x speedup using PySpark.
Text summarization and topic modeling system using LDA and NMF algorithms for document analysis.
Personal portfolio website with AI chatbot integration, dark mode, and responsive design.
Full-featured Twitter clone with real-time feeds, user authentication, and social interactions.
Predictive model for stock prices using LSTM networks, with a web interface for visualizations.
WhatsApp-based personal knowledge management system that captures messages, applies AI classification via OpenAI, and stores organized records in Notion.
A click-to-play platformer hidden behind the avatar on my homepage. Three hand-built levels, Koopas whose shells you can kick into other enemies, flyers, unstompable spikies, and warp pipes that charge a super jump. Runs on a pure, fully tested TypeScript engine with a canvas renderer, Web Audio sound effects, and touch controls.