nice work on this! i've been looking for something like this to manage my own docs.
one thing that caught my eye was the mention of 'proper stemming support' - can you elaborate on how you're handling stemming? are you using a specific library or rolling your own implementation? also, have you considered adding any sort of faceting/search filtering to the results?
Do the search results have document page numbers?
Yes this is great. I looked into stitching these together but always figured it would be a huge undertaking. Consider looking at TensorFlow for OCR which should be much better and maybe faster.