ML Module
Classify free-text into intents with an ML.NET model trained from your own labelled phrases.
Workflow Engine V 1.2.0Description
What this does
Train a multiclass text classifier from intents and example phrases you define directly on the task. The model is trained at deploy time, addressed by content hash in Azure Blob Storage, and served at runtime through a pooled prediction engine — so identical training data reuses the existing model and predictions stay sub-millisecond.
Predictions below a configurable score threshold are surfaced as the sentinel __not_recognized__, letting the workflow fall through to a fallback branch via SwitchTask.
What you can build
- Bot intent router for IVR or chat:
InputTask→MLNetTask→SwitchTaskto branch onbooking,cancel,complaint, … - Lightweight ticket classifier that categorises inbound emails or form submissions before fan-out
- Voice menu without DTMF — classify the speech-to-text result and branch on intent
- Triage gate that routes only
support/salestraffic into a paid downstream model - Spam / off-topic guard ahead of an expensive API call
See the setup guide for configuration and full task reference.