Note pipeline
Select a row to view the note text
A synthetic sample is loaded automatically. Click "View Note" to see the AI processing pipeline from original text to structured results, or upload your own CSV below. All bundled data is fully synthetic, no real patient information.
An LLM reads an unstructured patient note and pulls the clinical and histologic artifacts it finds (steatosis, ballooning, inflammation, and so on), then reports the fibrosis stage when one is present (with its scale, e.g. METAVIR / Ishak) and the disease the findings are consistent with (NASH, PBC, autoimmune hepatitis, alcoholic liver disease). Every call points back to the exact span it came from. Annotators review and mark each extraction to build ground truth, and that labeled data trains a smaller specialized model (SLM) to do the staging-and-attribution at scale.
This viewer is the review step: open a note, see what the model pulled and where, and flag what needs correcting.