TalkChat

Inspiration

We wanted to try out a more advanced language model, and this hackathon provided us with the perfect opportunity. We sought to provide a solution that may assist a person in discovering his/her/their/...issues on their own. Furthermore, we believe that chatting to other people is necessary, but this would provide a more comfortable interim solution for the consumer.

What it does

It uses scraped chat logs to train a sophisticated language model, specifically GPT-2, to produce responses based on the user's input text. No one else has access to the user's chats.

How we built it

The web application was built with flask, HTML, CSS, and JavaScript. GPT-2 was trained with the Python library gpt-2-simple, which is based on TensorFlow.

Challenges we ran into

We trained a BERT model on Google Colab for sentiment analysis of GPT-2 data, but owing to dependency difficulties, we were unable to load it locally.

Accomplishments that we're proud of

I am proud on my self that I have completed this project and able to make this chatting application and submit it to this hackathon.

What we learned

GPT-2 Architecture Flask, HTML, CSS, JavaScript, TensorFlow, Keras, HuggingFace

What's next for TalkChat

Training GPT-3 and gathering a lot more data so that the model can learn more complex patterns.

Built With

• css3
• flask
• gpt-2-simple
• html5
• javascript
• keras
• python
• tenserflow