SafeHaven

Empowering communities with real-time AI threat detection.

Comment

Inspiration

From our experience with our college sending out crime alerts, we wanted to build a system that provides more live feedback for users rather than waiting to receive an email about the aftermath. Oftentimes, these vague emails leave students stressed and confused. SafeHaven bridges this gap by allowing for real-time updates that follow a situation rather than sparsely sent emails.

What it does

SafeHaven is an AI-powered community security ecosystem that transforms passive surveillance into proactive protection. Unlike traditional security systems that merely record incidents for later review, SafeHaven actively monitors live environments to detect threats the moment they emerge. It does this via AI detection and user feedback to take notes of incidents, and provides tools for users to remain safe in these situations by using their location information and charting its relation to threats. Users also have access to a dashboard of threats in their area to always stay informed.

How we built it

SafeHaven was built with the NextJS framework, with Flask for routing. On top of this, Gemini API and YOLO worked in tandem to run a computer vision model to detect threats and verify their existence. On top of this, users and threats are stored in the Google Firebase and are used to keep track of how users interact with threat posts and save their location information and phone information. We used the Textbelt API to send real-time texts to users when new threats are sent to their dashboard as well. We also used Eleven Labs API to provide a real-time voice assistant to individuals who find themselves in these threatening situations, aiding them on a way to get out.

Challenges we ran into

One of the big issues we ran into was working on how to send text messages to users, which required heavy research and exploring multiple different options until we found one that worked for us. Another challenge we ran into was image upload in the carousel of images for threats, which required several implementation attempts.

Accomplishments that we're proud of

We are proud of our speech-to-speech AI implementations, as it was our first time enabling this, and we found that we were able to implement fairly swiftly with good results. Additionally, we were proud of our implementation of computer vision, as it was our first time as a team implementing this.

What we learned

We learned how to integrate several APIs into our backend while also learning how to use a computer vision model for a web development project for the first time. We also learned how to boundary box videos to look for specific information and how to then parse that information via the Gemini API for validity. These were some of the biggest new features we implemented.

What's next for SafeHaven

Next steps for SafeHaven include creating a mobile configuration for the website to make it more easily accessible for users and adding more voice features for users. We also plan on adding more features and more distinctions between the different threats beyond just labeling them as "generic threats." Additionally, we would want to implement real security camera footage, and the fight footage we gathered mainly serves as a placeholder. Additionally, the user thresholds for archiving an event would be higher, as the current threshold is only for demo purposes.

Share this project:

Updates