Privacy-Preserving, Discipline-Specific AI Writing Feedback: A Public Good for Under-Resourced Rural Districts

Team: Brian Pool, Jen Couch, Dan Studebaker, Sarah Hollingsworth

Overview: This project enhances four production open-source Moodle plugins to provide privacy-preserving, discipline-specific AI writing feedback for under-resourced rural schools. It operationalizes formative feedback principles in ELA, Science, and Social Studies across grades 5–12.

Subject Focus: Writing, Grade 5-12

Targeted Universalism Focus: Centers under-resourced rural students (57.4% free/reduced lunch) and learners with IEP/504 plans who lack access to specialized writing coaches or paid cloud-based AI tools. On-premise AI server infrastructure eliminates vendor subscription costs and prevents student data from leaving district servers, removing privacy and financial barriers for under-resourced districts.

Public Goods & Deliverables:

  • Enhanced Moodle Plugins: Updated open-source plugins (MooChat, MooProof, AI Check, AI Grade) published to Moodle.org under Apache 2.0/GPL.
  • Discipline-Specific Prompting Framework: Prompts grounding AI feedback in ELA knowledge integration, science argumentation (Toulmin), and social studies sourcing/corroboration.
  • Evaluation Dataset: A de-identified corpus of ~14,832 student writing samples across initial/final drafts, AI feedback, and outcome ratings under CC-BY-4.0.
  • Feedback Quality Benchmark & Deployment Guide: Open rubrics, test sets, and an IT deployment guide for rural districts operating on-premise LLMs (Ollama).

Return to see all grantees.