Building a Family Tree Back to 1700 as a Software Engineer

Most family trees stop at great-grandparents. Everything older is buried in 18th-century handwriting, scattered across Russian, Ukrainian, Lithuanian and Finnish archives, in scripts nobody in the family can read. In one month I took one line of my tree back to Peter the Great's census of 1710, eleven generations up, and found that my great-grandparents were seventh cousins. This is not an "ask ChatGPT about my ancestors" talk. It's an engineering story: a Claude Code harness of specialised agents (archive scouts, a paleographer, a refuter whose only job is to break every kinship claim), handwriting recognition running on a home GPU farm over hundreds of thousands of archive scans, and an evidence model where every link in the tree points to a document. I'll show where AI is brilliant (crawling catalogues, reading pre-reform cursive, never getting bored), where it lies confidently, and the guardrails that make the result trustworthy enough to put your grandmother's name on it.