Problem

Schools and departments receive repeated administrative questions about program differences, application rules, forms, and graduation requirements. Manual replies are repetitive and require checking multiple announcements.

Data source

The demo uses public documents and pages related to NTUT iFIRST, organized into Artificial Intelligence, Information Security, and Semiconductor knowledge collections.

System design

The flow uses document chunking, vector retrieval, source citations, and a Local Ollama LLM backend, with LINE as the user entry point.

Test questions

Questions cover program categories, application methods, course information, FAQ items, and source traceability. The test is not whether the answer sounds good, but whether it follows official materials.

Result

The demo shows LINE Q&A, document retrieval, source citations, and multiple knowledge collections, making it a practical starting point for education or admissions FAQ PoCs.

Limitations

Public data is not the same as a formal internal knowledge base. For production, the unit still needs to confirm current documents, permissions, update workflow, and error correction.

Next improvement

Add admin update workflow, usage records, more real question sets, and formal demo videos so the organization can evaluate ROI more easily.