AI solution design · Knowledge and RAG
Support knowledge copilot
A panel inside the ticket screen that suggests an answer from manuals, release notes and past tickets, for the agent to check and send.
Talk about something like thisThe situation
A support team answers customer questions using product manuals, release notes and years of closed tickets. The knowledge is spread across three tools, the manuals differ by version, and the best answer is often in a ticket someone solved two years ago.
How it works
Step by step
- 1
Index everything together
Manuals, release notes and closed tickets go into one search index, each tagged with its product and version.
- 2
Filter before searching
When a ticket opens, the search is limited to the product and version it concerns, so a reply about version 4 does not quote the version 2 manual.
- 3
Suggest a reply
The agent sees a draft answer, the passages it came from, and links to similar solved tickets.
- 4
The agent decides
Nothing is sent without the agent. They edit, accept or discard, and that choice is recorded.
- 5
Learn from the choices
Suggestions agents keep rejecting point to gaps or wrong documents, and show up in a weekly review.
What it is built from
- Azure AI Search
- One index across manuals, release notes and tickets, filtered by product and version
- Azure OpenAI
- Drafts the reply and summarises long ticket histories
- Ticketing system API
- Reads the ticket and shows the suggestion inside the screen
- .NET service
- Orchestrates the search and keeps the audit trail
Safeguards
- Drafts only: an agent always reviews before anything reaches a customer
- Version-aware, with a warning when two sources disagree
- Customer personal data is stripped before it reaches the model where it does not need to
- Accepted and rejected suggestions are logged for review
What you end up with
- A suggestion panel inside the existing ticket screen
- Summaries of long ticket threads
- A weekly list of knowledge gaps found from rejected suggestions
Where to start
Start with one product line and its most common ticket types. Measure how often agents accept the draft before widening it.
More in this group
Policy and procedure assistant
Staff ask a question in plain English and get an answer from the current policy, with the document and section alongside.
Read the design →Contract search and Q&A
Find renewal dates, notice periods and liability caps across hundreds of contracts, and ask questions with the clause shown in the answer.
Read the design →