YapBot

YapBot is your personal AI debate coach. Do you suck at arguing? Do you have that one friend that you never seem to win an argument against? Don't fear, YapBot is here! Ready, set, YAP!

Comment

Inspiration

We got into a debate about our hackathon idea. One thing lead to another, and we built YapBot!

What it does

  • Smart Argument Generation: Yapbot analyzes your debate topic and delivers well-structured arguments that are both compelling and effective.
  • Real-Time Suggestions: Get instant feedback and argument enhancements as you engage in debates!
  • Diverse Array of Styles: From Yoda to Peter Griffin Debate in the style of any character/personality of your choice!
  • Seamless Experience: Enjoy a user-friendly interface that makes accessing and utilizing Yapbot’s features effortless.
  • Real Time transcription: Get a real-time transcription of your debate with an option to export the transcription at any point of time.

How we built it

First we started out with experimenting with the Cohere API to have a working local script that we could use in a Flask backend that we needed to build for our hack. We then moved on to designing the REST API that our React frontend would ultimately communicate with. Once we had the API spec ironed out, our team split off into frontend and backend until we were done with YapBot.

Challenges we ran into

  • React/FE build systems - some of us were not too experienced with React, so getting the FE to look and feel good was a bit challenging.
  • Prompt engineering - it was quite difficult for us to settle on a good prompt to use with Cohere's LLM. After some trial and error, we settled on the one we are using on the backend.
  • Live audio streaming - figuring out how to live stream the audio from our computer microphone and get it transcribed to text was quite a challenge. We ended up using the Deepgram API with a websocket connection to ensure low latency live audio transcription.

Accomplishments that we're proud of

This was our first time working with the Cohere LLM API, so we were particularly proud of our prompt engineering that enabled the core functionality of YapBot.

What we learned

LLM hallucinations are very common without careful prompt engineering. We think that this is a particularly applicable skill now that LLMs are becoming more commonplace in all facets of society.

What's next for YapBot

  • Automatic voice classification/recognition (diarization)
  • Clone your voice + text-to-speech (TTS)
  • Debate transcript analysis (timeseries data analysis)

Built With

Share this project:

Updates