Initial feedback is that the UI on the webpage seems busted, or at the very least poorly-scaled. For example the "Celebrity Picks" photos are way too large and pixellated, and the "Recommended Lists" title text is hard to read against the background of Hillary Clinton's red jacket.
The UI also seems a bit too busy - hard to know where one should start the moment they land on the page. Particularly that it is recommending me a bunch of genres and celebrity picks before I've even given your model an indication of what I want. It'd be a lot more user friendly to either start with a simple explanation of "how it works" followed by a "start here" button, or simply ask the user for the last 3 books they read and then SHOW how it works by providing actual recommendations.
As someone who reads a bit (and who has a librarian partner who reads a tonne), this sort of app NEEDS integration with goodreads. We both maintain lists of what we've read, and what we want to read, on goodreads. If it was possible to just paste a link to my profile (i.e. give you the goodreads user id) and the page would say "of the books on your want to read list, you should read X next", or "why is Y on your to read-list? you wont like it", or even, "your 'want to read' list is missing Z!", I'd sign up in a heartbeat.
You could even scrape ratings from goodreads profiles to better train your model. I already rate books there (out of 5), so there is already more sentiment data there than you'd get by asking me to just list books as "like/hate".
As far as how book-genie operates, I put in "Sci-fi dogs" and it failed to recommend 'City' by Simak, but rather gave me a bunch of titles with 'Dog' in the title that aren't even sci-fi. To me this isn't really a good output, and tells me that your algorithm is quite shallow and isn't really recommending based on actual content.
Hope you keep going, as this is definitely a hole!