Hackathon CHATBOT

screenshot of model

Inspiration

ChatGPT. We wanted to make a bot that would talk to users similar to ChatGPT, especially since the model architecture is so interesting.

What it does

The chatbot is able to make conversation and can do things like crack jokes. We also built a simple GUI for the user to interact with the chatbot in a fluid manner.

How we built it

We built the actual chatbot and the GUI on python. Tensorflow and the keras library were used to build and train the chatbot, while Tkinter was used to build the GUI that the user interacts with. Both of these were done on python, because that is the language that we are most familiar with.

Challenges we ran into

Getting the model to compile without errors was very time consuming. For example, one layer accidentally had the wrong value typed into it and we spent 30 minutes trying to find the error. Numerous little typos were found, but they were all fixed. Another issue was figuring out how to save the model, which we figured out by using pickle. On the GUI side, we were struggling with the formatting, and eventually we were able to figure it out by placing each component separately. We also had to figure out how to actually import the model into the GUI, since originally the chat function was built on a separate file to test it. We solved it by moving the function into GUI file and changing the input to whatever the user input.

Accomplishments that we're proud of

We are very proud of the fact that we were able to build this whole chatbot + GUI system in less than a day. We are also glad that we were able to figure out Tkinter in this same time period, since none of us knew how to use Tkinter beforehand. Luckily, some of us had machine learning experience that could be applied to the chatbot. Finally, we are proud of the fact that we were able to bugfix everything

What we learned

We all feel that we learned a lot from this hackathon. All three of us learned Tkinter, which would be useful if we ever need to do graphical design in python again. Some of us also learned how to use python for the first time, since only one of our team members had some experience with python. A final thing we learned was in debugging, it is way better to let your teammates help you. Struggling by yourself for an hour is pretty dumb when your teammate debugged it in 2 minutes.

What's next for Hackathon CHATBOT

In the future, we are planning to expand the training data so the chatbot can have a wider variety of responses. We are also planning to improve the GUI, or move the chatbot onto a website for easier graphical design. On the website we can also save model responses to a database and have users rate how well the model did, so the model can learn using reinforcement learning.