Evaluate accuracy on your own documents
At SanadGah, instead of numeric claims, we evaluate processing accuracy on a sample of your organization's real documents — with defined criteria, transparent reporting, and a focus on Persian printed documents.
Evaluation method
A sample of your organization's Persian printed documents is collected, and the OCR output is compared against reference text. The final report includes the overall accuracy rate and a breakdown of errors.
Evaluation metrics
Evaluation is performed based on three metrics: Character Accuracy, Word Accuracy, and accuracy of key fields in contracts and invoices.
Error analysis
Errors are reported broken down by type: connected letters, Persian numerals, punctuation, and specialized terms — to identify which aspect requires improvement.
Continuous improvement
Linguistic post-processing (NLP) is applied to OCR output to correct common Persian errors. This layer is continuously improved with new evaluation data.
Sample report
After evaluation, a full report — including the accuracy rate, sample output, error analysis, and a deployment model recommendation — is made available to the organization.
Current limitations
The current focus is on Persian printed documents. Handwritten documents, low-quality historical archives, and texts with non-standard fonts may have lower accuracy and are on the product roadmap.