Introduction
Welcome to the manual of expert.ai Platform NL Flow.
NL Flow is the part of expert.ai Platform where users manage workflows.
It has two components:
- The Web application where users create and manage workflows based on the predictive language models designed with Platform's authoring application or with Studio.
- The runtime hosting published workflows as Web resources that any program can request through a simple REST API to analyze documents.
Creating and publishing a workflow means making models available for practical use.
A workflow does not necessarily coincide with a model: it can be as simple as a model or as complex as your analysis needs, with more models, parallel branches, pre-processors and post-processors taken from those already available, written in JavaScript or Python. The workflow can also have conditional paths and loops. Custom components—for example to integrate external services like generative models or perform use-case specific manipulation of the data—can be built using a Java SDK.
It is even possible to have multiple, independent, flows in the same workflow: all the flows receive the same input while the overall output of the workflow is the concatenation of the output of the individual flows.
This manual has two sections:
- How-to guide: all the things you can do with the Web application.
- Reference guide: the description of workflow building blocks and other definitions.
The REST API that allows you to use workflows programmatically is described in a separate manual.