In this (very casual) project with some friends, we try to make a natural language model to mimic the group chat between us and the friends in the project.

This project is still ongoing, and my role in it is to develop the whole ML framework. We started with the simplest possible approach (Bag of Words model) as a baseline. We saw that this did not produce great results, so we are trying other methods.

Currently we are trying the Continuous Bag of Words model, making a miniature transformer, and finetuning gpt-2.

Depending on the quality of previous models, we will try other approaches in the future (RNN, LSTM, Fine-tuning of other models, …). This is a (very fun!) exercise in teamwork, task delegation, NLP and deep learning in python.

• skills: Teamwork, NLP, Neural Networks, Software Design, pytorch
• status: ongoing