Hi, everybody. I am currently using the response_format OpenAI published to get structured data from a long text. But the quality is significantly less than what I get if I use the structured data module make.com provides for OpenAI. How can this be? Are they using something else then me? Here’s a sample of my json:
{
“model”: “gpt-4o”,
“temperature”: 0,
“messages”: [
{
“role”: “assistant”,
“content”: “Some description on what OpenAI is supposed to do.”
},
{
“role”: “user”,
“content”: “Here’s a text from somewhere. Analyse and extract detail information.\n\n\{{ $(‘Scrape Site’).item.json.extractor.result.richTextClean }}”
}
],
“response_format”: {
“type”: “json_schema”,
“json_schema”: {
“name”: “some_name”,
“strict”: true,
“schema”: {
“type”: “object”,
“strict”: true,
“properties”: {
“date”: {
“type”: “string”,
“description”: “description which date I’m looking for”
},
“name”: {
“type”: “string”,
“description”: “description of the names I’m looking for”,
“valueExamples”: [
“Peter”,
“Maria”
]
}, …
Am I using the API wrongly? Is something missing in my JSON? Any advice would be greatly appreciated!