Endpoint added for testing NLU intent classification for a specific release/speech-model (API-Docs)
August 8, 2025
Improvements
Implemented new NLU which uses Rasa Version 3
Context
We are upgrading the underlying Natural Language Understanding (NLU) model in RBA from Rasa V2 to Rasa V3. This upgrade introduces improvements in speech recognition accuracy, training speed, and security.
The new NLU model is compatible with existing speech assets, no modifications are required from customers. Internal testing has shown performance gains across all evaluation criteria.
Migration Plan
The migration will be executed in two phases:
Phase 1 – Optional Adoption (Starting August 6, 2025)
Customers will be able to select either the existing Rasa V2 NLU or the new Rasa V3 NLU when creating a new release or modifying an existing one. The new NLU will be the default selection, but customers may opt to continue using the old model temporarily.
Existing releases will remain unchanged during this phase.
Phase 2 – Automatic Migration (Starting September 15, 2025):
We will begin migrating all tenants and releases to the new NLU model. This process includes automated retraining using the existing speech assets.
No customer action is required.
No downtime is expected.
The migration will be gradual, and exact timing for each tenant or project cannot be predetermined.
The migration is expected to complete within approximately two weeks.
Once the migration is complete across all environments, the option to select between the old and new NLU models will be removed.
August 6, 2025
Bug Fixes
Fixed a bug where LLM intent descriptions in different languages were not taken into account.
Fixed a bug that caused the project import to miss redirecting after importing.
Fixed a bug in the Azure Marketplace post sales flow