Inspiration
Walking alone at night is a real worry, and most safety apps make it worse — they need everyone in your circle to install something, or they only react after things have already gone wrong. We wanted something built for our own community at FIU: a friend gets looped in with just a text link, and the app itself can tell "she's distracted" apart from "she's stopped moving and not responding."
What it does
FIU PawlinaWalk is a check-in companion with escalating alerts. Before a walk, you save a Primary and Emergency contact. Start a walk, set a check-in interval, and the app pings you to confirm you're okay. Miss a ping and it checks your GPS — still moving escalates gradually, stationary escalates faster. Alerts go: self check-in → Primary Contact → Emergency Contact → a simulated "would now contact emergency services" screen (fake, so the demo is never at risk of a real call). Contacts get a link with no app install — just your last location and a plain-English summary from Gemini instead of raw coordinates.
How we built it
-Frontend: React + TypeScript on Vercel -Backend: Python + FastAPI on Render, running the escalation logic and GPS checks -Database + Auth: Supabase (Postgres) -Location: simple polling, no websockets -AI: Gemini turns session data into a one- or two-sentence summary when an alert fires Three of us split the work: frontend, backend/escalation logic + Gemini, and the contact-facing page + the simulated final alert screen.
Challenges we ran into
Tuning the escalation logic so it felt reliable, not jumpy — deciding exactly when "stationary" should skip ahead versus just speed up the timer. Keeping the contact side truly zero-setup while polling for fresh location data. Making sure Gemini could never block an alert — if the call fails, the raw data still goes out.
Accomplishments that we're proud of
Contacts never install anything — just tap a link. Stationary detection actually changes escalation speed, not just alert wording. And a scared parent gets a clear sentence instead of a pin on a map.
What we learned
How to design a state machine that's fast enough to matter but calm enough not to cry wolf, and how a small, tightly-scoped AI call — structured input, short output, hard fallback — adds real value without becoming a liability.
What's next for FIU PawlinaWalk
Verified emergency services integration, voice-based check-ins, route-deviation detection, and expanding beyond FIU to more campuses across South Florida.
Log in or sign up for Devpost to join the conversation.