Building a Reliable Knowledge Graph Beyond GitHub Copilot

I have a general question about knowledge graphs. We're using GitHub copilot, and my coworkers are convinced that just letting it write its own markdown files and then somehow prompting it to read them is a substitute for structured knowledge. I'm trying to figure out how I can add some kind of supplemental implementation that proves that the knowledge graph can be a more reliable resource for particular things. I'm just not sure what the very smallest MVP step would look like.