ChatMIT

Interface
Loading answer
Chat

Inspiration

Last year when I arrived at MIT, I was a fish out of water. I was lost in the whirlpool of deadlines, club events, and registration. I struggled to complete problem sets on time and didn't even consider making connections and friends. Thankfully, I quickly met my next dorm neighbor, Hassan. This experienced third-year student knew everything about the courses I was taking and where to go to explore MIT. Without Hassan, I would have failed 8.01 because of my lack of knowledge about Professor Mohammad's office hours and review sessions. Without Hassan, I would have missed out on so many internship opportunities because of my lack of knowledge of applications and their deadlines. I was extremely grateful for Hassan, but I constantly thought about all the students just like me who did not have this resource. Enter ChatMIT: a chatbot trained on insider secrets and up-to-date data about MIT. Know struggling students can always have a resource they could rely on.

What it does

ChatMIT is a chatbot that prospective and current MIT students alike can use to answer their questions and curiosities about this immensely wonderful, yet confusing institute! Ask it about good study spots. Ask it where to find certain buildings. Ask it for student testimonials about UROPs. The possibilities are endless.

How we built it

ChatMIT is a large language model built off of ChatGPT3.5-Turbo. It is trained on data from MIT's blog posts, registration websites, club websites, academic resources, and more. The front-end interface is made using React.

Challenges we ran into

There were two main problems that our group ran into: first, training the GPT model; second, collecting immense troves of data. Training a GPT model proved to be a tall task in a 24-hour timeframe. However, our group managed to do so after hours of experimentation and trial-and-error. Collecting the data to train the model also was extremely time-consuming: MIT's public data is commonly regarded as "confusing" and "hard to navigate." Our team spent hours sifting through this data across the web, finally creating the finished database.

Accomplishments that we're proud of

We are proud of our unique, utilitarian idea that was actually implemented in a 24-hour timeframe. We had plenty of doubters, but also plenty of supporters - all of whom motivated us to persevere. Furthermore, we are super happy with the UI, as it has the color scheme of MIT, with Tim the Beaver offering the advice!