Seekr: A live, map-based treasure hunt for real-world events

Seekr is an innovative live, map-based treasure hunt designed for real-world events. The organizers create riddles and hide locations on a map. Teams, consisting of up to four members, solve the riddles using their smartphones. Upon solving a riddle, they locate the spot on the map, prove their arrival by taking a photo or scanning a QR code, and submit the evidence.

A local AI system generates hints for each team, allowing them to progress through the hunt. The first team to find the treasure is declared the winner, and the organizer monitors the progress of all teams in real-time. This treasure hunt is suitable for schools, campuses, festivals, team-building events, and any outdoor game organizer.

A demo video and the code repository can be accessed via the provided links. The AI model used for generating hints and verifying photos is Gemma (gemma4:e2b), an open-weight model that runs locally through Ollama. The AI compares each player's photo with the reference photo provided by the organizer, scoring it out of 100. A score of 85 or higher passes, while borderline scores are reviewed by the organizer.

The system includes safeguards to prevent hint leakage, and if the AI is slow or offline, the hunt automatically falls back to the organizer's own hints. The application is built using plain Node.js with SQLite and no build step, utilizing server-sent events for live updates. QR codes are generated and decoded by a custom code written by the developer.

The pixel-art interface is created using plain HTML, CSS, and JavaScript. The project has undergone rigorous testing, with over 100 automated tests and a full browser end-to-end run. One of the key advantages of Seekr is that it operates without relying on a cloud service. By running the AI locally on the organizer's laptop, there are no API fees, no usage limits, and player photos and locations remain within the venue.

This architecture also ensures that the application functions seamlessly during field trips or festivals with poor internet connectivity. Furthermore, the open nature of the AI model allows anyone to inspect, modify, or enhance the system, fostering a culture of open innovation and collaboration. In contrast, a closed API would have resulted in additional costs per photo, the transmission of every player's photo to a third party, and potential service disruptions whenever the internet or the provider experienced issues.