Apply for community grant: Academic project (gpu and storage)

#1
by ckharche - opened

Project Name: Next-Gen Curriculum Optimizer

Problem:

University course catalogs and prerequisite chains are incredibly complex. Students struggle to build optimal degree plans that match their career goals, often leading to delayed graduation or wasted credits.

Solution:

This is an open-source tool to solve that problem. It's an "AI Academic Advisor" with a React frontend and a Python (FastAPI) backend.

Students input their profile (completed courses, GPA, career interests like "AI/ML"). The backend then generates a complete, validated, semester-by-semester degree plan.

How Hugging Face Is Used (Resource Justification):

The HF ecosystem is critical for the "AI" part of this project.

  1. GPU (For Model Inference): The backend uses two HF model types:

    • Sentence Transformer (BAAI/bge-large-en-v1.5): I create vector embeddings for all course descriptions. This lets students find relevant electives semantically (e.g., searching for "game dev" finds "graphics" and "physics simulation" courses).

    • LLM (meta-llama/Llama-3.1-8B-Instruct): This model analyzes the student's profile against the complex graph of prerequisites to generate a reasoned, human-readable plan. The GPU is essential to make this inference fast enough for a live web app.

  2. Persistent Storage: This is needed to store the pre-processed university curriculum graph (as a .pkl file) and to cache the downloaded HF models, preventing re-downloads on startup.

This is an academic project that I started during my masters in Computer Science at Northeastern University & I want to see it complete now that I have graduated & working on finding a job while on F1-OPT. I am working on this with a distinguished professor who specializes in curriculum design. The goal is to deploy this as a free, open-source tool for my student community and create a framework other schools can adapt.

Sign up or log in to comment