Dead Drop
Hide a real object outdoors, hunt others' by riddle — an AI referee checks the find photo, never a meeting.
Project intro
Dead Drop is a mobile-first, real-world hide-and-seek network. A hider stashes a small object in a public place, describes only the vicinity (never the exact spot), and the app's AI referee writes a cryptic riddle for it. A hunter browses riddles near them, walks outside, finds the object, and photographs it; the referee compares the photo against the hider's description and issues a strict match or no-match verdict, and both sides score.
There are no feeds, no infinite scroll, and no meetups: the app never introduces players to each other and never arranges a meeting, and the exact spot is never published. It is for anyone who wants their neighbourhood to feel a little more like an adventure, and every screen is designed to end with the player going outside.
Key features include riddle generation with a deterministic safety scan that rejects addresses, coordinates, phone numbers and URLs, photo verification with a confidence floor (a model 'yes' below 0.72 is forced to no), a leaderboard, email plus Google/GitHub sign-in, and a local Ollama path so the referee can run on your own machine instead of sending photos to Google.
Vibe-coded evidence, in the author's own words on DEV (11 October 2026): 'Built with Cursor as the agentic IDE (Grok 4.7, high-effort planning mode) against a written master prompt.'
Honest limits, as the author states them: the hosted demo sends find photos to Google AI Studio so Gemma can referee them; seeded sample drops cannot be scored because nothing was physically hidden; and the drop-photos bucket is public by URL (paths are UUIDs and the UI does not publish them). GitHub stars: 0 at the 28 October 2026 check (12 commits, MIT licence); the live demo was opened and confirmed working on 28 October 2026.
Actions
Tech stack
Project info
- Project category
- Games
- Project status
- Online
- Collaboration intent
- None