A host creates a room and shares a link or QR code, and each teammate pitches an idea at their own privacy level, openly, with their name hidden, their idea hidden, or fully incognito. The moment the last person submits, AI takes the most distinctive element from every pitch and fuses them into four brand-new project concepts the whole team owns, crediting only the people who opted in.

• Live pitch feed over Supabase Realtime; the moment the last idea lands, AI fuses four fresh concepts (the host can also trigger early)
• Per-person privacy with two independent toggles, hide name and hide idea, enforced server-side so the browser never reads a raw pitch
• Gemini primary with an automatic Groq (Llama 3.3 70B) fallback, so a rate limit mid-event goes unnoticed

An AI-native air traffic control console that works real, live air traffic and grades its own AI against reality. It ingests live ADS-B transponder returns around any of 12 major world airports, pulls the current METAR, and runs an autonomous ATC decision core that sequences arrivals, allocates runways from live wind, monitors separation, assigns gates, and voices realistic VHF phraseology. When a flight commits to the approach the core locks its plan as a prediction, then grades it against the real landing.

• Autonomous ATC core: runway config from live wind, arrival sequencing by ETA, CPA separation monitoring with a 150s lookahead
• Self-grading AI scorecard: every locked prediction is checked against the real landing, with online learning that sharpens per-airport priors over time
• 3D and 2D TRACON scope in three.js, dead-reckoned at 60fps, with a full rules engine that runs 100% in the browser (no API keys)

Score any public GitHub profile 0 to 100 across languages, commit habits, repo quality, and README quality, tuned to a target role. Then close the gap: a Fix Kit drafts the missing README, LICENSE, and CI workflow for every weak repo, a roadmap shows the exact score gain of each fix, and a live badge, shareable reports, head-to-head compare, and five AI-suggested projects round out the loop. Free forever by design.

• Role-targeted scoring: languages by real byte counts, a day-by-hour commit heatmap, per-repo and README quality
• Fix Kit generates the actual missing pieces, an AI-drafted README, an MIT LICENSE, and a CI workflow, ready to commit
• Free-forever architecture: Gemini free tier with a built-in rules-engine fallback, so it never bills and never breaks

Friends in different timezones open the same link, type their name, pick a color, and drag the half-hours they are free across a 7-day grid. CalChat finds the overlap and shows it three ways: a heatmap of who is free when, a true week-grid calendar of overlap blocks positioned at their real times, and a recurring sweet-spots view that surfaces the day and hour of week that is most reliably open across many weeks. You can plan any future week, not just the current one, and every participant gets a personal note that everyone can read.

• All times stored as integer 15-minute UTC slots, so DST and half-hour-offset zones line up exactly when computing overlap
• React Three Fiber background of drifting calendar shards that ease into a stacked column every 14 seconds
• 22-color participant palette with server-validated uniqueness (HTTP 409 on collision)

Verified social platform that lets international students connect with home-country peers across U.S. universities via ".edu" verification. Includes mentor relationships from alumni and faculty, multi-region country-instance deployment, realtime messaging, and admin moderation.

• Full-stack across web and mobile with multi-region deployment
• Supabase Realtime messaging + role-based RLS for country-data isolation
• Progressive schema migrations + admin moderation systems

End-to-end content pipeline: ingests 75+ daily items across Hacker News, Reddit, RSS, and News API, runs a four-layer OpenAI prompt engineering stack with multi-weighted similarity detection and 78-entity tagging, and ships posts through a full-stack deployment.

• Multi-source trend aggregation with intelligent deduplication
• 45% reduction in irrelevant content noise via relevance scoring
• Deployed on Vercel, Render, and Supabase

Simple, fast task manager built to track daily tasks efficiently. Designed around low-friction capture and a clean daily view to keep focus on today.

• Lightweight UI with a focus on the current-day workflow
• Minimal dependencies, fast load, PWA-ready

Co-developed at the AI Summit Series Hackathon (Fetch.AI Innovation Lab, Runner's Up). A journaling app with mood and emotion tracking dashboards, multilingual agents, and weather-based insights that analyze environmental effects on mental health and recommend mood-elevating locations.

• Backend agents built with Python, Hugging Face, and OpenAI APIs
• Multilingual translation + empathetic response generation
• Mood analytics + personalized recommendations for teenage users