Sentinelle

Safe Navigation for Pedestrians

Comment

Zoom Link for judging (in case the other one doesn't work)

Yalam, Srikar is inviting you to a scheduled Zoom meeting.

Topic: TamuHack2026 Sentinelle Time: Jan 25, 2026 01:45 PM Central Time (US and Canada) Join Zoom Meeting https://tamu.zoom.us/j/3887180658?pwd=Fs7H2XYfji8IgZWfENNmsxmKTXB35Y.1&omn=95297348247

Meeting ID: 388 718 0658 Passcode: scG36NKJC

Inspiration

Being an off-campus student at Texas A&M, living without a car is challenging at times. I remember my first week of class, Google Maps said the fastest route was through a road that didn't have a sidewalk. Although eventually, I got used to living in College Station, last semester, I rediscovered this issue when going to the Rec Center, being navigated through an active work zone. I wanted an app which would cater to people like me without cars in a car centric environment.

What it does

Sentinelle is primarily a navigation system offering pedestrians a wide array of options ranging from speed to safety. It uses a Gemini powered routing system which observes the Google Street View images, hazards reported from users, crime data, and population densities to rate a variety of routing options.

How we built it

I built Sentinelle using React and Tailwind mobile-first interface. I integrated Mapbox GL JS, which is a free way to handle complex routing geometries. The backend was built with Python and FastAPI and is the bridge between the geospatial data and the AI engine. For the AI workflow, we use Google Gemini 2.5 Flash to analyze static street-view snapshots along the route gotten from Maps Api. For infrastructure, I deployed the application on Vultr Cloud Compute. I used one Ubuntu instance hosting the frontend via Nginx, and a separate instance running the Python backend.

Challenges we ran into

Initially, I wanted to use Vultr as a high core cpu system to have parallel calls to Gemini to observe 100s of images at once. Due to Vultr's rate limiting of users who put in less than $50 however, I had to move to a parallel system. This caused me to decide to analyze less paths for demo purposes.

Accomplishments that we're proud of

I am proud to make a function app which prototypes exactly what I've been looking for. I am glad I got hosting to work and code setup already for scalable compute.

What we learned

Building the app locally went well, but then had build errors when I tried to deploy. I had to navigate Nginx reverse proxy configurations and manage SSL certificates across two different subdomains to prevent "Mixed Content" errors. I also learned that safety is mostly subjective as a dark street isn't always dangerous, and a well-lit one isn't always safe. I had to teach the AI this and I thought it was an interesting nuance that could still be improved.

What's next for Sentinelle

In the future, Sentinelle will be able to take in even more data regarding routes, allowing users to submit their own experiences with certain streets. There will also be a walk-with-me mode, where Sentinelle listens for any dangerous sounds that occur during your walk, alerting a trusted friend or authorities if you don't respond to a vibration.

Built With

Share this project:

Updates