Inspiration
The inspiration stemmed from watching my younger cousin consistently have a bad experience when going to their advisor. They signed them up for the wrong classes and didn't know at all my cousins major.
What it does
It utilizes a local llm called phi3:mini from OLLlama to query and answer questions from the user regarding a specific student at UMBC. It searches through the UMBC data base and grabs student information, class information, and reviews, then answers the query according to the student information in the system. Which streamlines the effort of an advisor allowing them to always have the right answers for the student.
How we built it
It was built using the database given by the DoIT at UMD, Neo4j and multiple python libraries (streamlit, dotenv, langchain_community)
Challenges we ran into
The model had to be scaled down from originally experimenting with gemma3:12b and llama 3.2 vision because of limited capabilities a laptop has in terms of power
Accomplishments that we're proud of
Pieces all the data together in order for the LLM to use
What we learned
Learnt how to use Neo4j and langchain_community library
What's next for Advisor helper
In the future this project can turn into a Agentic AI model for advisors to get all the information of the student and sign them up, or drop them from class, all from a simple query.
Log in or sign up for Devpost to join the conversation.