Comment by carlesfe

I built a chatbot startup in 2015. It integrated with Whatsapp (which was possible at the time with some hacks), and had:

- Multimodality: Text/audio/images input and output. Integrated OCR.

- Connection with an asterisk server, it could send and receive voice phone calls! I used it to call for pizzas to a local place via whatsapp. This was prior to Google's famous demo calling a hairdresser to book a haircut.

- It understood humor and message sentiment, told jokes and sometimes even chimed in with a "haha" if somebody said something funny in a group chat or sent an appropriate gif reaction

- Memory (facts database)

- Useful features such as scheduling, polling, translations, image search, etc.

Regarding the tech, I used external models (Watson was pretty good at the time), plus classical NLP processing and symbolic reasoning that I learned in college.

Nobody understood the point of it (where's the GUI? how do I know what to ask it? customers asked) and I didn't make a single dime out of the project. I closed it a couple years later. Sometimes I wonder what could've been of it.