Hi Community,
I am currently working on implementing a custom LLM-based Virtual Agent topic to handle natural language queries related to CMDB CI relationships.
Requirement Overview:
- The LLM topic should accurately interpret user queries such as:
- “What servers host the Finance App?”
- “Show me upstream dependencies for Database X”
- The initial scope includes supporting the following CI types:
- Applications
- Application Services
- Servers (Linux/Windows)
- Databases
- The solution should also handle contextual variables, meaning:
- It should recognize CI names directly from user input
- Support minor typos and case-insensitive matching without requiring exact syntax
Ask:
I would appreciate guidance or best practices on:
- Designing LLM topics for CMDB relationship queries
- Handling CI name recognition effectively (especially with flexible input)
- Any recommended configurations, scripts, or patterns to improve accuracy and performance
Any insights, examples, or references would be greatly helpful.
Thank you in advance!
