Theodore Roosevelt Library Now Lets You Chat With an AI 'TR'

WHAT HAPPENED: A team working with the Theodore Roosevelt Presidential Library took a 300,000-record archive — letters, speeches, photographs, official papers — and turned it into something you can talk to. First, everything was scanned and run through OCR (software that reads text off images). Then AI added labels and summaries, which human archivists corrected through a custom review app. Only after that did the cleaned-up material become searchable.

The showpiece is "Talk to TR": a full-size digital human that looks and sounds like Theodore Roosevelt, standing in a museum space with realistic lighting and directional audio. When a visitor asks about something modern — say, a current political fight — the system doesn't guess. It uses a technique the authors call cross-era analogical grounding, which rephrases today's question into a documented historical parallel, so Roosevelt answers from what he actually said and did. A look-up step keeps every reply tied to real sources, and streaming keeps the conversation feeling natural rather than paused.

Making it survive a real museum floor was as hard as making it smart. The system includes watchdogs (automatic checks that catch failures), session isolation so one visitor's chat never bleeds into another's, and services that restart themselves. The result: it can run unattended, all day, for hundreds of strangers, without a technician hovering nearby.

The honest limitation: the authors don't claim a controlled test or benchmark. They report lessons from operating a public exhibit. That means the recipe is promising and copyable — for other museums, companies sitting on old documents, even families with boxes of letters — but it isn't proof the approach works everywhere. And the bottleneck is human: archivists must still verify the AI's transcriptions and labels, because a confident digital Roosevelt that gets a fact wrong is worse than no exhibit at all.