To warm up on the kind of work we will have to do, i say we start by creating a python chatterbot to see how things work out.
We don't need to code the entire thing ourselves (nor can we), so building this as an experiment is the first step.
This is like, the bare minimum. We need to start small.
What i think we should do is work together to create a dataset for it; basically a text file full of conversations that is converted into .json and given to the bot as training. What kind of conversation? Any conversation at all. What made Tay work so well is they fed it a huge amount of random inane twitter conversations so it picked up on the idioms, phrase structures and grammar that people used. So if you feel like it; just pastebin a bunch of anonymized conversations, chatlogs, whatever.
If you want to play around with it (less preferable to just helping with the dataset), you need the following
Download and install all of these then install the chat bot. Check the wiki for documentation, it will create a .db where you run the script containing all it's learned.