AI-Assisted Scaffolding for Instructor-in-the-Loop Community Chat
Location
Bent Corridor, Science Center
Document Type
Poster - Open Access
Start Date
5-1-2026 12:00 PM
End Date
5-1-2026 2:00 PM
Abstract
Asynchronous student-instructor communication channels, such as email and forum systems, are vital for supporting learning but frequently suffer from latency and scalability limitations. Learning Management Systems at universities (e.g. D2L, Canvas, Blackboard) do not usually have chat functionality. To include chats, instructors often rely on community platforms such as Discord, Slack, or Piazza. While these platforms reduce response delays and lower the barrier to student inquiry, they simultaneously increase instructor workload and create unsustainable pressure for immediate responses. Existing AI-based chat support systems attempt to address this scalability challenge through automated generation; however, these often result in impersonal interactions that erode the instructor’s pedagogical intent. In this ongoing work, we introduce an AI-assisted community chat framework designed for scaffolding pedagogy and Socratic methods. Using a Retrieval-Augmented Generation (RAG) system grounded in course materials provided by instructors, our approach employs constrained prompt structures to generate guided inquiries rather than direct answers. This helps preserve the instructor’s unique pedagogical tone. The framework supports multiple bot roles operating at varying levels of scaffolding and provides a centralized instructor-facing dashboard for document ingestion, real-time intervention, and pedagogical reflection. Its modular architecture is implemented through containerized services that isolate key components of the pipeline, including document storage and preprocessing, retrieval orchestration, and vector-based semantic search using systems such as MinIO for object storage and Qdrant for embedding retrieval. This setup enables instructors to retain full control over deployment, course content, system behavior, and student-facing interactions.
Keywords:
AI, Agent, Assistant, Education
Recommended Citation
Vu, Nam; Hoang, Nam; Fani, Logan; and Luong, Sam, "AI-Assisted Scaffolding for Instructor-in-the-Loop Community Chat" (2026). Research Symposium. 17.
https://digitalcommons.oberlin.edu/researchsymp/2026/posters/17
Major
Computer Science; Mathematics
Project Mentor(s)
Linh Ngo, West Chester University
Md (Amir) Amiruzzaman, West Chester University
2026
AI-Assisted Scaffolding for Instructor-in-the-Loop Community Chat
Bent Corridor, Science Center
Asynchronous student-instructor communication channels, such as email and forum systems, are vital for supporting learning but frequently suffer from latency and scalability limitations. Learning Management Systems at universities (e.g. D2L, Canvas, Blackboard) do not usually have chat functionality. To include chats, instructors often rely on community platforms such as Discord, Slack, or Piazza. While these platforms reduce response delays and lower the barrier to student inquiry, they simultaneously increase instructor workload and create unsustainable pressure for immediate responses. Existing AI-based chat support systems attempt to address this scalability challenge through automated generation; however, these often result in impersonal interactions that erode the instructor’s pedagogical intent. In this ongoing work, we introduce an AI-assisted community chat framework designed for scaffolding pedagogy and Socratic methods. Using a Retrieval-Augmented Generation (RAG) system grounded in course materials provided by instructors, our approach employs constrained prompt structures to generate guided inquiries rather than direct answers. This helps preserve the instructor’s unique pedagogical tone. The framework supports multiple bot roles operating at varying levels of scaffolding and provides a centralized instructor-facing dashboard for document ingestion, real-time intervention, and pedagogical reflection. Its modular architecture is implemented through containerized services that isolate key components of the pipeline, including document storage and preprocessing, retrieval orchestration, and vector-based semantic search using systems such as MinIO for object storage and Qdrant for embedding retrieval. This setup enables instructors to retain full control over deployment, course content, system behavior, and student-facing interactions.

Notes
Presenters: Nam Vu and Nam Hoang