Inspiration
Jonathan Waugh
What it does
Gives various metrics to evaluate politicians' statements during speeches and debates.
How we built it
Used Agno to build an agent pipeline, deployed app on Fly.io using FastAPI and React.
Challenges we ran into
Diarization (dividing speech up by who is speaking) is difficult and there aren't good APIs sponsored for it. So we had to roll with a tuned whisper model that integrated with pyannote. Even this wasn't that great.
Getting GPUs is hard - we had to use Google Cloud and request a particular quota.
Accomplishments that we're proud of
- Basic diarization
- Working (but ugly) truth meter
- Successful deployment
What we learned
- Separating speech by speakers is called diarization
- Diarization is hard
- Getting GPUs is hard
What's next for Waugh
- More metrics
- Automated pipeline (add your own videos)
Login: admin@example.com LWsdtPhhJW2NwcdayVRvu_5WoiYqjaZKgKlX2bq
Log in or sign up for Devpost to join the conversation.