Section 7: RAG over Product Data and FAQs
- Prepare product data, prices, policies and FAQs for search
- Managed file search in the Gemini API or OpenAI compared with running your own vector search
- Hybrid search that combines vector search with filters such as category and price range
- Say I do not know when there is no data instead of guessing a price
- Lab: a bot that answers product questions from a sample catalogue
Section 8: Lab: Updating Data Without Changing Code
- Keep product data in Google Sheets or a database that the team can edit
- Sync changes into the index automatically
- Keep prompts and standard replies outside the code
- Re-run a fixed set of test questions before switching on new data or prompts
Section 9: Lab: Handover to Admins
- Handover triggers: the customer asks for a person, the bot is unsure, or it is a complaint
- Pause the bot for that customer only and let admins reply through Chat in LINE Official Account Manager
- Alert the team with a conversation summary and give control back to the bot when the case is closed
- Lab: an admin handover system with a case summary
Section 10: Guardrails, PDPA and Security
- Defend against prompt injection and attempts to push the bot off topic
- Define what the bot must never promise, such as discounts or refunds outside policy
- PDPA: data collection notices, retention periods and customer rights
- Store API keys and the channel secret securely
Section 11: Lab: Deploying to Cloud Run and Operations
- Containerise with Docker and deploy the backend to Cloud Run
- Logging, monitoring and alerts when the bot fails
- Control LLM cost by choosing models, limiting length and caching answers
- Measure real conversations and keep refining prompts
Section 12: Workshop: LINE AI Chatbot Capstone
- Choose a sample business: an online shop, a clinic or after-sales service
- Assemble the full bot: memory, RAG, function calling and admin handover
- Test with a realistic question set and hard cases
- Demo it on your own LINE account with a launch plan