Hi, those results are outdated for dialogflow. The agent in dialogflow was not configured for being hybrid (rules + ML). After configuring it to hybrid, the results are pretty good.

This is the accuracy results with a test corpus of 50 intents, trained with 5 utterances per intent and tested with 5 sentences per intent (different than the utterances), and with dialogflow configured correctly. I can share the excel with all the sentences and results if needed.