🎓 Finnish Universities — Generative AI Course Benchmark

Scrapes course data from public Sisu/Kori APIs and identifies courses covering generative AI using keyword matching and gpt-4.1-mini evaluation.

Select universities in the sidebar and click ▶ Run analysis to start.
ℹ️ How this works

Data sources (Sisu/Kori — in app today)

UniversityAccess
Tampere (TUNI)GET /kori/api/course-unit-search + course detail
University of HelsinkiGET /kori/api/course-unit-search + course detail
Aalto UniversityGET /kori/api/course-unit-search + course detail
University of JyväskyläGET /kori/api/course-unit-search + course detail
LUT UniversityGET /kori/api/course-unit-search + course detail

Not in app yet (Peppi API pending): Turku, Oulu, Eastern Finland — see docs/peppi-research.md.

Pipeline

  1. Scrape — Tier A (broad) + Tier B (gen-AI) search terms; merge IDs; fetch details
  2. Keyword filter — regex on course text (Finnish + English gen-AI terms)
  3. AI filter — gpt-4.1-mini evaluates the full scraped pool; returns relevant, reason, and skill clusters

Azure OpenAI

Set AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT (OpenAI-compatible /openai/v1 base URL), and AZURE_OPENAI_DEPLOYMENT. Credentials are not stored in the UI or logged.

Caching & logs

Data directory: /app/cache/ (CACHE_DIR on Azure). Scrape cache: {uni}_raw.json; AI cache: ai_eval_cache.json. Logs: logs/benchmark.log (rotating). Clear scrape vs AI cache separately in the sidebar. See docs/storage.md.

Limitations

Results are indicative, not comprehensive: scraping uses public API search terms and detail fetches (not authenticated bulk exports), and generative-AI relevance is inferred by regex plus an AI model. Treat counts as a lower bound with possible false positives.

Downloaded PDF reports may not display correctly in Firefox’s built-in viewer (blank or black pages). Use Chrome, Edge, or Adobe Reader instead.