How should I structure old support tickets for a RAG-based AI customer support agent?
Hi everyone,
I’m working on a project where I want to build an AI agent for customer support. The idea is that customers can ask questions about technical issues such as SSH, IP addresses, DNS, VPS, Outlook, etc., and the LLM should help them diagnose and solve their problems.
I’m using my own knowledge base + RAG, but I’m still a beginner and I’m not sure what the best way is to structure my data for retrieval.
I already have some old support tickets that I’d like to add to the knowledge base. These tickets usually contain:
- The customer’s initial problem/question
- A conversation between the customer and the human support agent
- Troubleshooting steps
- The final diagnosis
- The solution that was applied
For example, if a customer previously had an SSH connection problem and the support agent solved it by identifying a specific configuration/firewall issue, I’d like the RAG system to retrieve that previous case when the AI encounters a similar problem, so the LLM can use the previous solution to help the new customer.
My question is: how should I transform and structure these old support tickets before putting them into the RAG?
Should I keep the conversations as they are, or should I transform each ticket into something more structured, for example:
- Problem / symptoms
- Environment
- Diagnostic steps
- Root cause
- Solution
- Verification
- Similar scenarios
- Keywords / metadata
And how should I handle chunking these tickets so that the RAG retrieves useful parts without losing the context of the original conversation?
I’d really appreciate advice on how you would structure this kind of knowledge base, especially if you’ve built a RAG system for customer/technical support before.
Thanks!