A speed project generating text completions with an LSTM trained on the Mueller report. I used data prepared by Factbase with a little extra preprocessing to remove headings and footnotes. It's nowhere near perfect but you can find it at mueller-report.txt. Then I trained a network using this port of char-rnn by Sherjil Ozair and loaded it with ml5.js.
It doesn't work on mobile right now, due to some bugs with either ml5.js or TensorFlow.js.