Non-invasive Generative AI in Moodle via RPA and Retrieval-Augmented Generation

We present the design and deployment of TA BOT, a generative-AI teaching assistant that participates in Moodle discussion forums to sustain activity and keep replies aligned with course material. The system runs externally to the LMS using scheduled Selenium-based RPA and a retrieval pipeline built from lecture slides, where each slide is embedded and indexed in a vector store for contextual grounding. At runtime, a synthesized query retrieves top-k candidate slides that condition the model’s response. TA BOT operates under a dedicated Non-Editing Teacher account and follows a delayed-response strategy, intervening only after a thread has been inactive for at least 48 h so that students have the opportunity to contribute first. We report an engineering case study in a software engineering course, combining implementation-level process evidence with a thread-level quantitative analysis of forum logs. The analysis summarizes overall and pre/post deployment activity, including the proportion of discussions in which the BOT intervened, median response delay relative to the last human post, the rate at which dormant threads were revived, and the proportion of interventions that attracted subsequent human replies. Results show that TA BOT can be integrated non-invasively, operate reliably at scale, and help re-activate inactive threads without displacing human participation, contributing a practical architecture and lightweight, log-driven evaluation approach for constrained LMS settings.