Clearly not context aware: typing "Wonder Woman and Superman" gives:

"Woman" may be insensitive, use Person, Friend, Pal, Folk, Individual instead "Superman" may be insensitive, use Titan instead

There is a lot of work done in the area of linguistic analysis and word disambiguation, and none of it it's trivial. So it will be a big leap moving from a toy project to something of actual use. Integrating with an existing ML api for disambiguation could vastly decrease the amount of false positives. Idilia, BabelNet/BabelFly and LingPipe come to mind.