Integration
Ragas evaluation export
The evaluation export is JSON with question, ground_truth, and contexts. Those names match the fields a Ragas-style sample usually needs. You still build the sample and run the metric in your environment.
Field mapping
questionis the generated question.ground_truthis the expected answer, or the canonical answer when no separate expected answer was stored.contextsis a list. FAQai puts the item context in it, or the source chunk text when the item has a chunk id and no context string.
There is no answer from your production model in this file. Faithfulness and similar metrics need a response from the system you are testing. You supply that response. FAQai does not.
Illustrative mapping
Synthetic. This shows the export fields, not a completed Ragas result.
{
"question": "How long is the return window?",
"ground_truth": "30 days from delivery.",
"contexts": ["Items can be returned within 30 days of delivery."]
}Confirm the current Ragas sample schema before you rename fields. FAQai does not pin a Ragas version and does not claim every metric accepts this file with no adapter. Evaluation export requires a paid plan. The generation steps are in create a RAG evaluation dataset from a PDF.