Our summary
A helpdesk receives more than customer questions: automatic replies, failed-delivery notices and unwanted mail arrive too. Freek Van der Herten explains how his helpdesk app, There There, uses Jev to identify these messages. Laravel’s programming tools connect the app to this outside AI service.
Mail headers are labels added by mail servers. Some identify a message without AI. Jev checks the remaining questions together, using the subject, sender and shortened message text. It gives each answer a probability, a number showing how likely it thinks the answer is. Code turns that number into a yes-or-no result.
Later rules read saved results instead of asking Jev again. The app uses a 75% cutoff for automatic replies and a 90% cutoff for delivery failures and spam. It saves the probabilities too, so changed cutoffs can be compared. Errors do not stop the mail arriving, though shortened text leaves later details unchecked.
Key takeaways
- Check the labels added by the mail server first, then ask Jev the remaining questions together.
- Choose a cutoff for each question: the example uses 75% for automatic replies and 90% for delivery failures and spam. Save the probabilities so you can revisit those choices.
- Limit the text sent for checking and let mail processing continue if the check fails; a missing answer should not stop the inbox.