1. For a competent ML researcher, this is trivial to build using VQA + MCTS + standard message passing.
2. MARL techniques with attention-based communication solved this years ago
3. It does not seem very "viral" or income-generating. I know this is premature at this point, but without charging users for the service, is it reasonable to expect to make money off of this?