I Asked Gemini to Read 300-Year-Old Portuguese Parish Records (mariofilho.com)

A recent experiment evaluated the ability of AI language models, specifically Gemini 3.5 and 3.6 Flash, to extract genealogical data from 300-year-old Portuguese parish records. The task involved identifying baptism, marriage, and death records from digitized images while preserving the structure and relationships within the data. The results showed that both AI models could extract information like names, dates, and parental relationships with reasonable accuracy, but older records proved more challenging due to legibility and historical writing styles. This work is significant for the AI/ML community as it highlights the capabilities and limitations of current models in processing historical documents, which are essential for genealogical research. By demonstrating the models’ ability to handle complex tasks such as expanding abbreviations and normalizing historic names, the study showcases the potential of AI in transforming non-searchable archives into structured, usable data. However, challenges remain, such as varying performance based on record age, encouraging ongoing innovation in AI-driven text extraction and historical analysis tools.