A software engineer used AI to search centuries of historical archives and surfaced a forgotten meteorite report, three lost rhinoceroses and several unrecorded volcanic eruptions. The project was inspired by historian Benjamin Breen's October 1st, 2026 post, which used AI to search digitized Dutch East India Company records. Breen found a 1615 ship's journal describing sailors on Mauritius who "caught many tortoises, dodos," a new eyewitness record of the extinct dodo that had been sitting in plain sight for four centuries.
The Dutch East India Company dominated the spice trade and ran ports across Asia from 1602 until it was dissolved in 1799. The GLOBALISE project has turned millions of its handwritten pages into searchable text. Working with Claude Code, the author built a research pipeline that combined AI models to search multiple historical mysteries in parallel and automate tedious manual steps.
The main challenge was that historical spelling was inconsistent, and text recognition errors were common. The 17th-century Dutch spelled "rhinoceros" in roughly fifteen different ways, so keyword search would miss most relevant passages. The solution was to convert each passage into a mathematical representation of its meaning, allowing searches by subject rather than exact wording. A process that would have taken a person about 70 years ran through the Dutch East India Company archive in a single overnight session.
Every AI-generated claim had to be verified against an original document, since AI models can be confidently wrong. The approach aims to add small, checkable data points to niche historical fields.
Source: Hacker News · Summarized by HeadlinesBriefing