Inspiration

Marina Mogilko suggested helping her with inbound messages that she doesn't have bandwidth to handle atm

What it does

We've built a simple AI classification tool to sort through DMs and comments to identify problems and identify prospective leads. Then we engineered prompts for each classification

How we built it

We used Facebook and YouTube APIs to access comments and DMs, huggingface models to classify user messages and ChatGPT 3.5 to generate user responses.

Challenges we ran into

Facebook APIs and testing environment doesn't work. Classification didn't yield good results without models being fine-tuned

Accomplishments that we're proud of

Made the user message classifier work with a training dataset

What we learned

Facebook api sacks

What's next for Waterbear

Test it in real life, make it a fully-functioning product

Built With

  • ipywidgets
  • next.js
  • supabase
  • transformers
Share this project:

Updates