Document Processing / IBM watsonx Orchestrate
Make the trust loop visible.
Builders need to see what a document extractor found, check uncertain results against the source, and see whether their changes improved the output. I worked on Classifier and Extractor specifications and led the design of Accuracy Evaluation in IBM watsonx Orchestrate.
The work started with the platform, not the dashboard.
The workflow identifies the document type, extracts its fields and tables, supports human review, and compares the results with a reference.
I joined Cate on the design team in 2025. I helped adapt earlier concepts to the newer agent model, delivered interaction designs and specifications for Document Classifier and Extractor, and contributed shared canvas and component patterns.
I also worked with adjacent teams on shared document-review patterns.
Parallel feature arc
Classify, extract, review, then evaluate.
I worked on document setup and human review alongside the broader canvas work. Accuracy Evaluation came later, adding a way to compare extracted fields with reference values.
Document Processing feature arc
-
01 · Classify
Establish the document type
Suggested classes make the system recommendation visible before the workflow moves forward.
-
02 · Extract
Shape the information model
Field setup connects the source document to the structure the workflow needs.
-
03 · Review
Put uncertainty in context
Review keeps uncertain output and its source close enough for a person to resolve the difference.
-
Later phase · Evaluate
Make quality inspectable
Accuracy Evaluation compares extracted and reference values field by field.
Classify
Extract
Review
Evaluate
The screens use fictional sample data. The values shown are examples, not measured product results.
Trust comes from a workflow people can inspect.
Document Extractor and Accuracy Evaluation were released in July 2026.
Return to the Canvas work