ABBYY FineReader -- I don't have the specific version in front of me unfortunately.

In addition to the structured text we're currently serving through the API, we also have 300DPI color scans and per-word coordinates and confidence scores, so there's a lot more we can do with the OCR data that isn't exposed yet.

For those docs that require more finesse (This is OCR :D), was there a massive manual effort involved?

Nice question. That's what I want to know.