Winners
Five teams, two tracks, one weekend. Congratulations to the builders who pushed Claude AI the furthest.
Build anything
The projects that pushed creativity, engineering depth, and Claude usage.
EagleEye
by Claude Mythos
AI-assisted emergency dispatch that watches highway camera feeds and triages incidents in seconds.
EagleEye monitors Columbus highway cameras and uses Claude's vision to detect crashes and hazards in real time, generate structured incident summaries, estimate severity, and recommend the right emergency response package — compressing the gap between detection, triage, and dispatch.
React/Vite + Express/TS, ODOT camera proxy, scenario engine, multi-provider analyzer with mock fallback for demo resilience.
Tokenlens
by Crtl Alt Elite
Paste any GitHub repo URL → instantly see exactly how much that codebase spends on LLM API calls.
TokenLens scans every file, detects LLM calls across OpenAI, Anthropic, Google, Cohere, Mistral, LangChain, LlamaIndex and more, then projects costs across 20+ models. Live streaming results, token estimation by task type (RAG, summarization, coding, classification), plus a Claude Haiku recommender that suggests cheaper model swaps with rationale.
FastAPI + React + Tailwind (~6.2k LOC), 14-model pricing catalog, NDJSON streaming, SQLite caching.
Pilot
by DGX Disciples
A macOS agent that drives your iPhone for you — Claude vision + synthetic taps through iPhone Mirroring.
SwiftUI menubar app + Python daemon that uses Claude's vision to read your phone screen, plans actions, and injects synthetic taps and keystrokes through Apple's iPhone Mirroring window. Users write short YAML "skills"; after 3 clean runs Pilot freezes the workflow into a compiled local replay (~30s cloud loop → ~2s).
~10k+ LOC, MCP server, 6-layer memory (catalog, runtime store, vector index, hint cache, compiled snapshots, usage backplane), APScheduler.
For a better grid
Solutions tackling environmental and energy challenges with AI.
Greenify
by Awesome Dragons
Tell your house "I'm leaving" or "I'm cooking" — Claude reorganizes your energy use to match.
A smart home agent that takes plain-English intent and automatically adjusts devices to cut energy use, while protecting essentials. Every run reports both energy saved (kW) and dollar savings, with a monthly view tying decisions back to the bill. Safety-first: high-impact devices first, essentials untouched, full transparency on what changed and why.
FastAPI backend (1300-line agent, hybrid LLM + rules planner, schema-validated JSON plans, real smart-plug seam), React 19 + Three Fiber 3D dollhouse frontend.
EnergyWatch
by Forecast Fellas
AI dispatch for home solar — when to sell, when to charge, when to pull, explained in plain English.
Helps homeowners with solar + battery storage decide when to sell back to the grid, charge their battery, or draw from it. Pulls real Open-Meteo forecasts for solar generation and asks Claude Sonnet to reason over pricing signals, battery state, and the 7-day forecast — then writes natural-language advisories like "Tuesday looks overcast — pre-charge tonight at off-peak rates to cover the shortfall."
React + Node/Express + SQLite, bcrypt auth + Gmail verification + JWT, 3,288-line Recharts dashboard, customizable strategy modes, live power-flow diagram.
Huge thanks to every team who shipped this weekend, and to our judges from NVIDIA, Meta, Google, IGS Energy, AWS, and Vertiv for their time and feedback.
See you next year. 🧡