evil kermit

• 

  Proof of concept

Inspiration

seeing that our schedules were super busy and work was piling up, we found it hard to reply our friends and loved ones in time

What it does

automate replies to them based on our texting patterns and behaviours

How we built it

telegram bot api in python to register users, telethon for obtaining user-related data

Challenges we ran into

kept getting rate-limited from telegram's api due to extensive testing process inaccurate responses from trained model

Accomplishments that we're proud of

streaminlining every process into a telegram bot without the use of a middlewares such as web ui or other interfaces

What we learned

LLMs are not easy to work with, especially

What's next for evil kermit

improve our training methods/models

Built With

• gpt-2
• jupyter-notebook
• python-telegram-bot-api
• telegram
• telethon