Projects

deepThorin

Summary

Deep Thorin is a Deep Learning and A.I. project to mimic esports personality Thorin's unique tweet style.

The bot uses a multi-layer Recurrent Neural Network trained on tens of thousands of tweets and articles. It is written in python using tensorflow.

The tweeting is done manually due to the complexity of formatting esports tweets leading to a larger than expected error rate if I just tweeted out the output of what the bot is "thinking".

Thorin was chosen due to him having a large corpus of material (lots of tweets / data), as well as a unique style and genre which poses a greater challenge for any A.I. project. The challenge makes this project extra rewarding and fun!

The bot can be prompted with initial words, and it will try to complete the sentence, or it can just provide a 'random' sentence based on what it has learned.

This project is largely based on code and research done by Andrej Karpathy.

jmete3@gmail.com
© James Mete 2017