SBU Sense - AI-Powered Course Discovery Platform

Inspiration

As Stony Brook University students, we've all experienced the frustration of course planning. With thousands of courses, hundreds of professors, and complex degree requirements, finding the right classes can feel overwhelming. Traditional course catalogs are static and don't consider your academic history, interests, or goals. We wanted to create a solution that makes course discovery intelligent, personalized, and data-driven.

SBU Sense transforms course planning from guesswork into a strategic, informed decision-making process powered by real student data and AI.

What It Does

SBU Sense is a comprehensive course discovery and recommendation platform that helps Stony Brook University students:

Intelligent Course Search

Semantic Search: Natural language queries like "machine learning" or "organic chemistry" find relevant courses instantly
Advanced Filtering: Filter by subject, SBC requirements, professors, course level, letter grade distribution, and more
Real-Time Statistics: View A-ratio, student enrollment, and grade distributions for every course

AI-Powered Recommendations

Personalized Suggestions: Describe your academic interests and goals, and get tailored course recommendations
Transcript Analysis: Upload your PDF transcript to automatically extract your course history
Smart Matching: AI analyzes your background, prerequisites, and progression to suggest the perfect courses
GPA Boost Calculator: See estimated GPA impact for each recommended course

Data-Driven Insights

Leaderboards: Discover top-performing courses and professors by A-ratio
Professor Profiles: Compare instructors across courses with detailed statistics
Course Progression: Understand prerequisite chains and recommended next courses
SBC Tracking: Automatically track which Stony Brook Curriculum requirements you've fulfilled

How We Built It

Backend Architecture

FastAPI: High-performance Python API with async support
AI Integration:
- OpenAI GPT-4o-mini for advanced reasoning
- Pinecone vector database for semantic search
Data Processing:
- Scraped and processed 100,000+ student grade entries over 10 years
- PDF transcript parsing using pdfplumber
- Course bulletin scraping and indexing

Frontend Architecture

React + TypeScript: Modern, type-safe UI development
Vite: Lightning-fast build tool and dev server
Tailwind CSS: Utility-first styling for rapid UI development
Recharts: Data visualization for statistics

Key Features Implementation

Vector Search:
- Embedded course descriptions using OpenAI embeddings
- Semantic similarity search via Pinecone
- RAG (Retrieval-Augmented Generation) for context-aware recommendations
Transcript Processing:
- PDF parsing to extract course codes and grades
- Automatic course history extraction
- GPA calculation and progression analysis
Recommendation Engine:
- Multi-stage filtering and ranking
- Prerequisite checking
- Course progression analysis
- Professor rating aggregation
Real-Time Search:
- Keyword matching across course codes and titles
- Multi-criteria filtering with AND/OR modes
- Sortable results by A-ratio, enrollment, etc.

Challenges We Ran Into

Data Quality & Processing

Challenge: Processing 100,000+ grade entries with inconsistent formatting
Solution: Built robust data cleaning pipelines and validation checks

PDF Transcript Parsing

Challenge: Transcripts have varied formats and layouts
Solution: Implemented flexible parsing with regex patterns and fallback strategies

AI Response Reliability

Challenge: Ensuring AI recommendations only include valid courses
Solution: Multi-layer validation, course code verification, and fallback mechanisms

Performance Optimization

Challenge: Fast search across large datasets
Solution: In-memory indexing, LRU caching, and efficient data structures

Vector Search Integration

Challenge: Setting up and managing Pinecone vector database
Solution: Automated index creation, dimension detection, and error handling

Accomplishments We're Proud Of

Comprehensive Data Coverage: Successfully processed and indexed 100,000+ student records spanning 10 years

Advanced Functionality:

PDF transcript analysis
GPA boost estimation
Course progression tracking
Prerequisite validation

Real-World Impact: Built something that solves a genuine problem for thousands of SBU students

What We Learned

RAG Architecture: How to effectively combine vector search with LLMs for context-aware recommendations
Data Engineering: Processing and cleaning large-scale educational datasets
PDF Processing: Techniques for extracting structured data from unstructured documents
AI Prompt Engineering: Crafting prompts that produce reliable, structured outputs
Performance Optimization: Caching strategies and efficient data structures for fast queries
Full-Stack Integration: Seamlessly connecting React frontend with FastAPI backend