Hey everyone,
I’m working on a project involving a massive number of API endpoints (around 15,000 from the Microsoft Graph API), many of which depend on or interact with each other. Given the sheer size, it’s impossible to simply feed all of these endpoints directly into an LLM. The challenge is figuring out how to ensure that an LLM can fully understand all the relevant endpoints and their relationships in order to provide accurate results, especially for complex, multi-step queries.
Question: How would you approach feeding an LLM all this information? Not just the endpoints themselves, but also dependencies, required parameters, and use cases? Are there any best practices for setting up such a system?
I’m thinking about using a graph database (like Neo4j) to represent the API endpoints and their relationships. Has anyone worked with this kind of solution or have other suggestions?