Dormsurf

Logo

Inspiration

As out-of-state first year and transfer students, both of us recognized the challenges of adjusting to a new institution and the need to step outside our comfort zones to form meaningful connections with our peers. Unlike high schools and community colleges, the sheer size of a large public university like Georgia Tech means that it's less likely to see the same familiar faces everywhere you go. That's why we wanted to create a platform that seamlessly connected college students with like-minded peers. After doing some background research, we realized just how prevalent the issue of loneliness was for college students in today's day and age, and wanted to create something that truly made a difference in addressing the issue.

What it does

Dormsurf allows college students whose roommates are off-campus to find other students who have an extra space in their dorm on weekends or days when their roommates are also off-campus. It works by creating unique user profiles that take into consideration the student's own dorm, basic information, and roommate preferences to match them to other like-minded students who live close to them within Georgia Tech's campus and have an available space for the selected timeframe. The app also provides a map that helps users navigate to their weekend stay and also provides a custom messaging platform to coordinate rooming logistics.

How we built it

Using Cursor's agentic coding platform, we built the backend with Flask and frontend with ReactJS. As for the functionality, Meta's Muse Voice Transcribe converts users' audio responses to questions about their hobbies and habits into text. Meta's Muse Spark 1.3 then constructs a structured Pydantic object with fields representing various attributes as well as a rich personal bio that gets displayed on the UI. The structured output is then embedded as a 32-dimensional vector, and cosine similarity is used to determine matching profiles. These profiles are then filtered based on important characteristics e.g. year in college, gender, etc. and re-evaluated using Muse Spark 1.3 to ensure a more accurate ranking of profile matches. For data storage, MongoDB is used to store user credential data while Pinecone.io vector database stores the embeddings. Finally, the frontend and backend are deployed using Vercel.

Challenges we ran into

One of the key challenges we ran into was accurately displaying a map that could help users navigate to their weekend dorm. The initial map produced by Grok 4.7 was crudely designed and directions were represented by arbitrary lines rather than actual roads. After researching open-source APIs, we instructed a different model, Claude Fable 5.1, to use Leaflet to render the map, which yielded much more positive results. Additionally, we also had several issues with deployment that were primarily caused by merge conflicts, which we resolved by looking through the codebase to identify the correct changes.

Accomplishments that we're proud of

We are proud of making a multi-dimensional app that not only matches college students with weekend roommates, but also provides navigation and messaging functionality. This would make the app so much more seamless to use as separate applications need not be used for getting to the dorm or coordinating stay logistics.

What we learned

We learned the importance of delegating tasks based on specialized skillsets, for both humans and AI. For example, I was better at high-level system design, while Ben was more experienced in deploying the website and debugging issues in the codebase. Similarly, Grok 4.7 was more cost-effective at developing most of the platform, while Fable 5.1 was better at aesthetic design, such as the map implementation.

What's next for Dormsurf

We hope to expand Dormsurf's capabilities by scaling it to other colleges in the country. As we gain users, we also plan to consider partnerships with universities to sponsor our product in order to help promote a greater sense of community within their college.