Local LLMs + RAG + BKT for privacy-preserving adaptive learning + faculty tutors
Key Questions
What is the core technology behind this highlight?
It integrates local Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and Bayesian Knowledge Tracing (BKT) for privacy-preserving adaptive learning with faculty tutors. This combination enables practical Intelligent Tutoring Systems (ITS) that run on-device, addressing limitations of cloud-based solutions.
What results have been achieved with Moodle integration?
Moodle fusion has demonstrated a +15% success rate improvement in adaptive learning environments. This supports scalable, privacy-focused educational tools.
What gains were confirmed by the UPenn Python tutor?
The UPenn Python tutor, using LLM+ML sequencing, confirmed 6-9 month learning gains for novice programmers. It highlights the effectiveness of adaptive tutoring for skill acquisition.
What is the Sandpiper app and its purpose?
The Sandpiper app scales analysis of tutoring transcripts to identify effective teaching moves. Developed in connection with efforts like the National Tutoring Observatory's open-source tool, it enables conversation analytics at scale for improving teaching practices.
What is the current status and thesis focus of this project?
The project is in the developing stage. Its thesis emphasizes ed-RAG/BKT-LLM frameworks, on-device MOOCs, interaction and equity scaling benchmarks, and conversation analytics.
Moodle fusion +15% success; faculty ADAPT tutor; UPenn Python tutor LLM+ML sequencing confirmed 6-9mo gains for novices; Sandpiper app scales tutoring transcript analysis for teaching moves; new on AI tutors as 'teacher-software' reinvention and Khan-simulated learners for math teacher training. Addresses cloud gaps with practical ITS. Thesis: ed-RAG/BKT-LLM, on-device MOOCs, interaction/equity scaling benchmarks, conversation analytics, teacher sims.