Answers should be traceable.
Learners need help finding and interpreting course material without treating a chatbot as an authority on grades, personal matters, or unsupported facts. The prototype keeps the knowledge boundary visible.
Approved material
A curated JSON corpus of syllabus excerpts, selected slides, handouts, reflection prompts, and case texts. Source IDs and locations stay attached to each passage.
Relevant passages
A local lexical retrieval step selects matching excerpts. If the evidence is weak, the agent says so instead of filling the gap.
Learning support
Without an API key, the prototype displays original excerpts. With a server-side key, it can generate a cited answer and falls back to excerpts if citation checks fail.
Try the decision logic.
This is an illustrative walkthrough, not a live connection to the course agent. Choose a question type to see how the prototype is designed to respond.
Show the evidence first
The learner sees relevant original passages and their source locations. A generative answer is optional and must cite the selected passages.
A research prototype, not a deployed campus service.
The supplied application runs on a local Python server at 127.0.0.1. It has no login, role-based access, conversation database, or secure multiuser deployment. Before course use, it needs source review, evaluation cases, institutional privacy review, and an authenticated hosting design.
Human oversight
Questions beyond the corpus are directed to the teacher or official channels. The system does not grade student work.
Test beyond fluency
Measure source accuracy, justified refusal, useful learning prompts, and independent learner decisions.
Deployment gate
Review source rights, accuracy, privacy, costs, access control, and operations before making the service available to learners.