ProfIQ — AI-Powered Professor Analytics Platform
Search and compare professors using student-review sentiment, recurring themes, and recommendations based on similar review profiles.
- React
- Django REST
- NLP
- scikit-learn
Make reviews easier to compare
ProfIQ brings professor search, review analytics, and comparison into one interface, helping students explore patterns across reviews before choosing a class.
From individual reviews to a clearer profile
Professor profiles combine sentiment summaries, recurring review themes, and similar-professor recommendations. A comparison view places those signals side by side for up to four professors at once.
Build a recoverable data pipeline
The professor catalog and review corpus are collected from the RateMyProfessors GraphQL API. Because collection runs long enough to get interrupted, the pipeline paginates through results, checkpoints progress, and deduplicates records, so a dropped connection resumes cleanly instead of re-fetching everything or creating duplicate entries.
Choose models for the running application
The live app scores every review with a VADER-based rule engine extended with a professor-review lexicon (76.4% test accuracy, text only) and, when available, a TF-IDF/logistic-regression classifier (80.2% accuracy on a larger 52,445-review test set) as a second signal. It then recommends similar professors using MiniLM embeddings — a separate task measured by how often its top-5 matches share a department (~7x more often than random), not by sentiment accuracy. DistilBERT was evaluated offline only. It reached the highest accuracy at 85.7%, but it was fine-tuned on just 8,000 reviews and scored on the smaller 7,427-review test set that VADER used, so its numbers aren't directly comparable to the classifier's. Transformer inference is also too slow to run inline on every API request, so it remained a comparison point rather than a live upgrade.