Plan: Identify the scenarios in which users might use your application. Define the actions and relevant information that needs to be recognized. Think about the schema that would be best suited for your application. Learn more about planning your model here.
Teach: Define your schema from scratch or with the aid of Duet templates. Upload your unlabeled dataset or use a publicly available dataset. Teaching is an iterative process that guides the user through building the model incrementally. To teach your model, you need to explore your data as you build your schema using the editable schema capabilities. Add a few labels to each category in your schema as you grow it. Address the conflicts that appear by selecting the most relevant features from the suggested machine learning features. Keep iterating on your model while monitoring the quality metric associated with each category until you are satisfied with your model. Then move to test your model on some unseen documents. Test your model by typing a query and obtaining a prediction result; if you're not happy, continue to teach your model.
Publish: Deploy and publish a new version of your model. You can integrate your model with GoogleSheet or Zendesk or you can consume the model through a REST endpoint whose URL is provided after the model is deployed.