It would be super helpful to have a large (well, large by our standards) corpus of good job examples.
These should be individually vetted and approved to ensure that they meet modern practice.
The examples don't necessarily need to be part of apollo - they could live in the docsite, for example we already have lots of job examples there. The problem is that the average quality is dangerously low. The API docs also have examples, which are probably better quality on average, but have very little composition.
For our most popular adaptors, we need at least one example of every function, but we also need examples of composition.
Later, we may want to embed this corpus into the vector DB to assist RAG. Or use it to train our own model
An obvious concern here is that the quality of the examples will degrade over time as our standards and APIs inevitably change
It would be super helpful to have a large (well, large by our standards) corpus of good job examples.
These should be individually vetted and approved to ensure that they meet modern practice.
The examples don't necessarily need to be part of apollo - they could live in the docsite, for example we already have lots of job examples there. The problem is that the average quality is dangerously low. The API docs also have examples, which are probably better quality on average, but have very little composition.
For our most popular adaptors, we need at least one example of every function, but we also need examples of composition.
Later, we may want to embed this corpus into the vector DB to assist RAG. Or use it to train our own model
An obvious concern here is that the quality of the examples will degrade over time as our standards and APIs inevitably change