Educational Knowledge Graph & Automated Summarization

Research Problem

Can AI generate pedagogically accurate, multi-granularity academic summaries — per lesson, week, month and subject — by aggregating heterogeneous data sources, while ensuring privacy-safe content that excludes identifiable information about other students?

Technological Innovation

An automated summary service orchestrates data collection across lesson content, session transcripts, activity submission results, attendance records and teacher notes. Time-window aggregation produces structured summaries that route through sensitivity-gated AI infrastructure, and a teacher-approval workflow (generated → teacher review → approved → distributed) ensures human validation before any AI-generated content reaches parents or students.

Why This Is Hard

Multi-granularity academic summarization is a novel NLP challenge with no existing benchmarks. Aggregating transcripts, scores, attendance and notes into coherent narratives requires cross-modal summarization at the frontier of AI research. A hallucinated progress claim would destroy institutional trust irreparably, and privacy-safe summarization must discuss one student’s performance without leaking details about classmates — entity separation that current LLMs handle inconsistently.

Success Criteria

Greater than 85% teacher approval of generated summaries without major edits, measured through edit tracking in the approval workflow. Summarization is deliberately conservative — understating rather than overstating progress — with a fact-checking layer against source data.