Back to CoursePack AIAG-013 Data Infrastructure

Data & AI Infrastructure Policy

This policy outlines how uploaded course files are processed once, compressed into structured course prediction profiles, and governed across student accounts.

1. Uploaded Materials Purpose & Processing Pipeline

Uploaded materials (lecture slides, syllabi, quizzes, past midterms, past finals) are processed through a modular ingestion pipeline:

  1. Text extraction and SHA-256 document hash deduplication.
  2. Structural document chunking without silent truncation.
  3. Per-chunk topic classification and question extraction.
  4. Deterministic backend score aggregation for topic probabilities.
  5. Generating a compressed Course Prediction Profile.

2. Raw Materials Privacy vs Shared Derived Intelligence

Raw Uploaded Files (Private)

Raw PDFs, PPTX slides, and images are strictly private to the student who uploaded them. They are never shared, displayed, or served to other users.

Derived Prediction Profile (Shared)

Only anonymized, compressed topic probability distributions (e.g. "Proof by induction has 87% historical appearance") are shared among students taking the same course section.

3. Data Retention & Version Invalidation

- Material Invalidation: Uploading new course materials or deleting materials increments the material_version, which automatically invalidates old cached outputs.
- Profile Invalidation: Updating backend prediction models increments prediction_profile_version, forcing a transparent recalculation for future predictions.
- Cache Lifetime: Cached outputs remain active until underlying course materials or prompt versions change.

4. Account and Material Deletion

Students can delete uploaded files at any time directly through the dashboard interface. To request full account and profile data purging:

  • Go to Account Settings > Delete Account.
  • Or email privacy@coursepack.ai with your registered student email address.
5. Mandatory AI Prediction Disclaimer (免责声明)

AI exam predictions, topic difficulty rankings, and countdown calendars are probabilistic study recommendations. AI predictions do NOT guarantee actual exam content or student final grades (AI 预测不保证真实考试内容或成绩). Students must study all required course topics.