CallBot

Inspiration

Our own life - we've been super distracted on many conference calls taking notes and not participating in the call itself, because humans are not good multitaskers. Now that we have a bot that transcribes the entire call and emails us the most important points minutes after it took place we can fully focus on actively participating in the conference call, bring as much value as possible and review the notes afterwards.

What it does

It's a bot that dials into conference calls, like on Google Hangouts or Zoom.us, transcribes them, then does a little bit of NLP on the result and emails participants the notes from that conference call.

How I built it

The core of the project is based on Twilio which is being used to dial into the conference calls using a US-based phone number and a PIN. Twilio interacts with our backend built in Node.js which coordinates recording and transcription of the call as well as takes care of emailing the result to the participants. Speech recognition, participant identification and natural language processing is built on Google Cloud Services. Because of the length of some calls we had to make the entire call pipeline completely asynchronous and independent of other services.

Challenges I ran into

Wifi at the venue :) Also connecting all the different services our hack depends on did turn out to be challenging especially for debugging.

Accomplishments that I'm proud of

Over the course of 16 or so hours we managed to prototype the solution to the full extent. Maybe without a proper business model however as we know from history this isn't going to stop us conquering the world ;) We successfully built, configured and connected as many as 4 different API providers along with our custom backend to make CallBot work.

What's next for CallBot

This is not another throw-away hackathon project. We plan to release it as a working service once we clean up the code and spend some time working on an actual business model for this. Because some of the dependencies we're using are paid for, we do need to at least cover the costs right from the get-go. We firmly believe we can have first real world users of this sometime in January of 2020 so about a month after the hackathon concludes. We got high hopes of getting covered on TechCrunch :)