CASE 01 — INTERNAL AI ASSISTANT
Answers grounded in internal documents
According to the employee handbook, follow these steps:
01 — WHAT IT DOES
The assistant looks for relevant information in Google Drive and other approved sources, answers from that material, and lets people inspect the original document.
02 — USE CASES
It can support employees internally or answer customers on a public website.
01 Internal · Slack
Example questionHow do I claim travel expenses?
Guide employees through the expense policy and required documents.
@Knowledge Assistant How do I claim travel expenses?
Select “Travel” in the expense system, enter the route and amount, and submit by month end. Attach a route screenshot if there is no receipt.
▤ Expense policy (sample), section 4Thank you. I have submitted it.
02 Customer · Website
Example questionWhen can I use this service?
Answer only from approved public FAQs. Keep internal documents and personal data out of the index.
When can I use this service?
The service is available weekdays from 9:00 to 18:00. Please check the FAQ for holiday hours.
Source: FAQ / Opening hours03 Internal · Customer support
Example questionHow should I answer this question?
Suggest a draft based on past responses and FAQs, with a person reviewing it before sending.
Can I change my delivery date to next week?
Thank you for asking. If your order has not shipped, you can change the delivery date from your account page. Here are the steps.
We choose the chat interface together: Slack, a dedicated chat, a site guide, or an existing tool, based on its users and operations.
Discuss your project ↗03 — DIFFERENCE
Two chats answer the same question. The key difference is whether the system searches internal documents first.
How do I apply for parental leave?
How do I apply for parental leave?
04 — TECHNOLOGY
It combines internal document search with generative AI using RAG (Retrieval Augmented Generation).
A generative model does not know your company handbook. For each question, the assistant retrieves relevant passages from approved documents, then forms an answer with links so people can verify the original.
Convert documents and questions into vectors with the same model so similar meaning can be found even when wording differs.
Connect ingestion, the search index, chat API, and AI generation while keeping retrieval and answer generation separate.