Campus Compass

Inspiration

Campus Compass started when we found ourselves overwhelmed at EAF (Enormous Activity Fair) trying to decide which clubs to join. With so many options, we wanted a tool that could instantly match students’ interests to campus organizations in a clear, personalized way.

What it does

Campus Compass takes a student’s resume or typed interests and recommends the most relevant UCLA clubs. It combines AI-powered semantic search with natural-language summaries to give students concise, human-like explanations of why each club fits them.

How we built it

We built a Streamlit app as the UI, used AWS Bedrock with Amazon Titan models for embeddings and text generation, and stored vectors in FAISS for fast similarity search. We scraped the UCLA club website in pieces, exported the data as JSON, and then wrote a script to convert it into a clean CSV for our pipeline. To summarize the architecture of the project is as follows: We used a Streamlit front-end, converted resumes into embeddings with AWS Bedrock’s Titan Embed model and stored in a FAISS vector database. When a query is made, the system retrieves the most relevant clubs and feeds them to AWS Bedrock’s Titan Text Express LLM, which generates personalized recommendations displayed back in Streamlit.

Challenges we ran into

Scraping the UCLA site was difficult because the pages were inconsistent and had to be exported as a JSON file first. As a result, our prototype only has around 200 clubs from UCLA. Formatting the model’s outputs was also tricky; we spent time iterating prompts to stop it from adding extra bullets or skipping clubs. We also planned for the model to “learn” about the user over time but couldn’t implement that yet.

Accomplishments that we're proud of

We turned a messy problem into a working prototype that really helps students. We successfully integrated retrieval-augmented generation with a friendly interface and a curated dataset. This was especially impactful as we did not have any previous experience with AWS software.

What we learned

We learned the importance of clean data pipelines, prompt engineering for consistent AI output, and how to stitch together multiple AWS services to create a seamless experience.

What's next for Campus Compass

We plan to extend Campus Compass to more universities, improve output consistency, and add real personalization so the app remembers student preferences across sessions. We also want to have the model scrape the club websites and be able to add events to the user's calendar.