armqna - Tensorflow based question+answer tool

[tong]

Slightly OT / another experiment: https://github.com/tong/armory_qna

A tensorflow based tool for questioning the armory wiki in human language.
Idea is to run this in a workflow to automatically search for answers in questions extracted from issue/pr comments (probably labeled with "question"), and post any findings as comment.

The workflow part isn't implemented yet, but i've added a manually triggerable workflow for testing.
See: https://github.com/tong/armory_qna/actions/runs/3696813126/jobs/6261018782

You can also run this locally.
If you ommit the --question parameter it runs in interactive mode, which allows to make several questions without loading the model again (takes some time).

Note: This is just a quick hack!
I think to be useful the wiki markdown files need to be formatted to plain text and also skip some files which don't have any information (footer.md, ...).
And probably add other sources as knowledge input (api docs, issues/pr/discussions, ...).