People are going missing in foreign countries and unfamiliar places.
Relevant to our lives as college students who go out with friends.
Wanted a solution that works automatically, no action needed from the person in danger.
Important to keep track of people intoxicated.
What it does
Real-time friend tracking on a live map.
Passively monitors gait (how you are walking), ambient sound (is your sound environment going from loud to quiet), and proximity (how close you are to friends).
AI reasons over all signals and automatically alerts your group with an explanation of why someone may be in danger.
Progressive alert tiers: All Clear, Heads Up, Concern, Urgent, Emergency.
Help button, pinnable safe spots, notification log, end-of-night summary, past rallies history.
How we built it
Mobile-first web app in vanilla HTML, CSS, JavaScript.
Node.js + Express backend, Supabase for auth and data.
Leaflet for maps.
AI safety layer sends all signals to Claude, which reasons holistically and returns a safety tier + notification blurb.
Hardcoded AI responses for demo due to API cost.
Challenges we ran into
Designing the AI detection logic ( when to actually fire a notification without false positives).
Deciding whether to use a scoring formula vs. letting the AI reason holistically (went with holistic).
Actually paying for API credits.
Accomplishments that we're proud of
AI reasoning layer that explains why someone was flagged, not just that they were.
Built a full working prototype with a clean, intuitive UI.
Designed a system that works passively. No action needed from the person in danger.
Lots of interactive features like hydration reminders, ordering Uber, contacting emergency contact.
What's next for Rally
Convert to native mobile app (React Native / Flutter) for real sensor access.
Train a dedicated model on real gait data.
Add escalation system such as personalized follow-ups to closest friend, then emergency contacts if no one responds.
Log in or sign up for Devpost to join the conversation.