When I use an API, I employ ts-node and axios without any libraries. I always inform GPT in the prompt about the desired data and the expected data format for the response. For example:
Return
```json
{"items":
[
{
"1": "City",
"2": "City",
"3": "City",
"4": "City",
...
}
]
}
Out of 10,000 requests, 20% of them are incorrect, and I have to repeat them. I validate the returned response using the Joi library.
Perhaps it would be better to validate the data at the prompt submission level. Is it too much that 20% of the data is not validated?