Govern tutor knowledge through teacher-curated document corpora with RAG and document-driven adaptation¶

Strategy · All strategies
Evidence · no claims cited

Description¶

The Teacher Interaction Layer lets educators delimit what the virtual tutor may know by curating an approved corpus. "Educators can curate a corpus of materials (e.g., textbook excerpts, lecture notes, worksheets, institutional policies, or open educational resources) that delimit the knowledge boundary of the system." These materials ground responses through retrieval-augmented generation at query time or support lightweight fine-tuning where policy permits, aligning explanations with the exact terminology and sequencing used in class.

Design Implications¶

Context¶

Requirements¶

• Teacher time to curate and maintain the document corpus
• Institutional policy permitting the chosen adaptation mechanism
• Tracking of corpus versions and update workflows for auditability

Constraints¶

• Fine-tuning or continual adaptation is described as feasible only when permitted by institutional policy

Target Learners¶

• students in institutionally governed school deployments

Target Learning Goals¶

• factual consistency and curriculum-aligned explanations grounded in approved sources

Affordances¶

• Elevate Three Stratum Genai Avatar Tutor Framework

Related Strategies¶

• Curating Resources
• Adopt AI incrementally in waves from simple to complex tasks, on free-tier tools, with mandatory multi-step human review before any document is issued
• Use a non-generative AI chatbot with a pre-programmed, human-curated knowledge base to answer student questions with pre-approved responses
• Validate wiki-grounded course tutors through classroom deployment, harder questions, larger corpora, and hybrid or graph RAG comparison arms

Examples¶

-

Key Sources¶

• Lorenzo Stacchio, Michele Giordano, Daniele Berardini, Primo Zingaretti, Emanuele Frontoni. (2026). ELEVATE: Designing Human-Centered GenAI Virtual Tutors for Scalable and Inclusive Education. arXiv preprint. https://arxiv.org/abs/2606.30662