APIs
Our data pipeline powers APIs that are useful for different purposes including data exploration, labelling, analytics, and downstream NLU tasks.
Predict#
API description
Receives an utterance and returns ranked list of intents with similarity score
Once the NLU Model is trained within studio, it's immediately ready to serve requests via the API. See more details on its dedicated page. Training the NLU model also has the side-effect of triggering a job which classifies all unlabeled data using this newly trained model.
Query#
API description
Allows querying unlabeled data using trained model
For example, search for a specific keywords, or get a breakdown of how many times an intent was matched using the latest trained NLU model, in a given time period.
It's possible to compose queries using multiple predicates in order to find relevant unlabeled examples (and their surrouding context, if you are dealing with conversational data).
info
Unlabeled data is indexed in multiple ways:
- Using a full text search for traditional querying based on different predicates.
- Using trained NLU model, each utterance is classified and the output probability distribution of intents is available for conversational analytics at scale
Recommend#
API description
Powers the suggestions and semantic search features within studio
note
This API is coming soon. Let us know of any interesting use case you might have.