Turing team member here. You can also try the demo at the following link which operates at pixel level for images (no meta-data which is how similar systems like search engines work). Another cool thing is that the model does inherent OCR without explicit training or inference set-up for OCR. https://turing.microsoft.com/bletchley

Happy to answer any questions.

What does this have to do with decoding German U-Boat transmissions?