Exact Sciences
ESOCR, patient intake OCR
An OCR pipeline for patient intake forms, handwritten and printed, validated in a web UI as the data is extracted.
40%
higher field-level accuracy on the hardest handwriting
- Intake formprinted and handwritten
- Nanonets OCRtrained on synthetic sets
- Express APIFirebase
- Next.js UIfield-by-field validation
The problem
Intake forms arrive with a mix of printed and handwritten fields, and the handwriting was where extraction failed most.
Stack
Next.js, Express, Nanonets, Firebase, OCR
View the repositoryWhat I built
- 1An OCR pipeline on Nanonets for handwritten and printed fields.
- 2Synthetic training sets that improved the Nanonets model on the hardest handwriting samples.
- 3A Next.js UI where staff validate each field as it is extracted, backed by an Express API on Firebase.