Back
Challenge
Latin-language Vatican reports, handwritten in the 15th century, presented significant issues for transcription — including archaic abbreviations, challenging penmanship, and decorative watermarks.
Solution
Osiris-AI delivered:
- Custom HTR models trained on idiosyncratic Latin scripts, achieving ~90% accuracy
- Palaeographic expertise to identify abbreviations and language variants
- Image preprocessing tools to remove watermarks
- Output in standardised formats to aid interpretation
Impact
This project demonstrated that AI transcription can be applied even to highly stylised manuscript sources. The results are being used for private theological research.