Inspiration
I oftentimes stay late into the night, & find it very uncomfortable walking around despite being in my own school. Thus, I thought of making a map app to keep us all safe. One in two women also feel uncomfortable, so I feel that this is a cause I can devote myself to with pride, not to mention other disadvantageous people
What it does
It retrieves criminal data for regions along your route & maps the safest way possible. It uses RAG & FSL to reliably ascertain criminal data with how uncertain police report data can be.
How I built it
I used a vite/npm frontend, overlayed with the mapbox api, & multiple geocoder apis. I implemented a middleware to better streamline the communication of such humongous volumes of data, using express to partially decode into formats appropriate for my backend. It also handled the algorithm for finding the path, courtesy to my one & only reliable teammate. The backend is then a lot of python with analytical & AI tools, where I used hamming codes from libraries to normalize address data, then refit them into the FSL AI model I personally trained. It is then reinforced further with RAG with a few reliable sources.
Challenges I ran into
As a noob to AI, training one was really tough, especially when I was trying for two
Accomplishments that I'm proud of
I'm proud to make a functional AI model from scratch, as well as making the frontend look as nice as it does (I'm usually horrendous at UI/UX). Map apis have also been hard for me so this felt relieving. I'm also proud that I can still do full stack, not easily, but still able to, after a long break from web development
What I learned
I learned that my project management skills are a far cry away from good, as I found myself always switching around how I handled data that costed me time. Likewise, I feel like I should be firm & logical on my priorities, as I still aim too high.
What's next for soteria
Creating a reliable mciroservice architecture to centralize criminal activity data. Would need to get in touch with local police departments for this. I also want to implement reinforced learning from continuous user feedback, probably stored in a vectorized mongodb database. The performance is also something I need to work on, as it is quite inefficient as it is now. Last but not least, introducing other elements to better identify danger, such as whether places tend to be crowded or not, during what time of day. This was a really fun time for me & I'm glad to have done this project.
Log in or sign up for Devpost to join the conversation.