AcademicAdventure

Inspiration

While the UMass schedule builder is great, is doesn't have the capacity to help students plan for the long term.

What it does

Our application considers all course prerequisites in order to give students a path to take the classes they want without having to compare many different documents.

How we built it

We took an open source Qwen language model and connected it to a small RAG document base, allowing the model to determine the prerequisites of a given class.

Challenges we ran into

We had a difficult time getting reliable results from the small language model we used. At first we spent a long time trying to get the language model to output into a format convenient for parsing, but we ended up transitioning to a reliable representation closer to plain language. That improved reliability, and we were able to construct a parser to get the format into our final desired representation.

Accomplishments that we're proud of

No one in our group had built a RAG model before, although some of us were familiar with the concept. We're proud we were able to scrape together our own version of the RAG algorithm (intentionally without relying on external libraries for RAG specifically).

What we learned

We learned that although LLM's can be quite powerful, even when small, they are very unreliable and require a lot of configuration and adjustments to get even semi-consistent results. It also became clear to us that small models were quite incapable of even basic reasoning.

What's next for AcademicAdventure

Right now all we have is a proof of concept. We want to leverage the existing RAG language model for more general questions, we want to be able to deploy this model to help students, and we want to add additional tools to ensure students can fulfill their requirements (major requirements, honors requirements, etc)

Built With

• fastapi
• htmx
• llm
• python
• rag