Rosheta
Medical Prescription OCR
An end-to-end research pipeline for turning difficult handwritten medical prescriptions into usable text across English and Arabic workflows.
Explore the project
- Built a medical-prescription OCR system with three pipelines for preprocessing, English, and Arabic, using 8 or more models across classification, handwritten OCR, autocorrection, and spelling correction.
- Fine-tuned T5 for autocorrection, applied transfer learning to OCR models, and trained YOLO on custom data for line segmentation, creating an end-to-end research pipeline for difficult handwritten prescription text.
- Evaluated NLP spelling-correction algorithms under limited-data constraints, balancing model accuracy with practical dataset limitations.

