We build OCR, starting with the parts that usually get left out.
A lot of OCR works well on clean English documents. It gets harder with other scripts, old scans, handwriting, unusual layouts and languages with little training data. That's the part we're interested in.
Runemic is a small company building OCR models, software and scanning tools. We're starting with multilingual OCR and an API, then moving into document processing and digitization.
Rune + Numeric
Rune is writing and symbols. Numeric is data and computation. Runemic is where the two meet: turning documents into data.
Our logo is the rune ᚱ (the letter R) inside a scanner's frame. The corners are data points, and the line across it is the scan.
Models first
Everything else depends on the models, so we start there. Next come the API and the scanner app, then digitization services and scanning hardware.
How we work
Open weights where we can
We publish models, with their evaluation code, when the licence, the training data and the economics allow it.
Measure things
We benchmark on real documents and report accuracy for each script and document type in RuneBench, including where we're weak.
Don't pretend OCR is perfect
OCR makes mistakes. You should be able to see them, measure them and correct them, so we return confidence scores and keep the output easy to check.
Your documents stay yours
We don't train on what you send us, and we offer ways to run the models on your own machines.
Work with us
Working with documents, looking after an archive, or want to help? Write to us.
hello@runemic.com