Tomorrow @ Brown

Inspiration

We realized that although there are really cool events happening on campus every day, it can be hard to find out about them unless you're on the right email lists. We recognized an opportunity to aggregate and filter this diffuse set of events and present it to members of the Brown community through a search engine interface.

What it does

We accept a query from the user and return the three most relevant emails regarding on-campus events! The more verbose the prompt, the more accurate the responses will be.

How we built it

We accept a query from the user describing the kind of event they're looking for on-campus. Then, using a pre-computed corpus of emails scraped via the Gmail API, we use the TF-IDF algorithm and the cosine similarity score to rank emails by their relevance to the user's query. Finally, we return the top three-ranked emails to the user.

Challenges we ran into

We initially sought to use the Open-AI API in order to call GPT 3.5 Turbo when selecting the most relevant emails, but the small context window proved prohibitive without first aggressively shrinking the search space. Although the TF-IDF algorithm was originally meant to be merely a pre-processing step, we were happy enough with the accuracy of its rankings that we decided to eliminate the LLM altogether in order to reduce our response time.

Accomplishments that we're proud of

We ideated and completed the project all within 24 hours, which we're proud of!

What we learned

This was the first Hackathon for the majority of our team members, so we learned a lot about how to work together collaboratively under tight deadlines.

What's next for Tomorrow @ Brown

We see many opportunities to boost both the speed and accuracy of our returns, using pre-processing techniques as well as using LLM embeddings to compute the cosine similarity score.

Built With

• python
• react
• tailwind