Spolling Bree is a version of the NYT Spelling Bee game that only accepts fake words that are plausible.
Originally plausibility was rated by a simple statistical model trained on trigrams from every English word.
The current model is machine learning based, and uses a character based RNN with GRUs trained locally on English words.
Play the live demo at bee.jminjie.com.
To deploy at localhost:8998
, run python3 server.py debug