eTA | Devpost

Inspiration

We noticed a lot of stress among students around midterm season and wanted to utilize our programming skills to support them both mentally and academically. Our implementation was profoundly inspired by Jerry Xu's Simply Python Chatbot repository, which was built on a different framework called Keras. Through this project, we hoped to build a platform where students can freely reach out and find help whenever needed.

What it does

Students can communicate their feelings, seek academic advice, or say anything else that is on their mind to the eTA. The eTA will respond with words of encouragement, point to helpful resources relevant to the student's coursework, or even make light conversation.

How we built it

Our team used python as the main programming language including various frameworks, such as PyTorch for machine learning and Tkinter for the GUI. The machine learning model was trained by a manually produced dataset by considering possible user inputs and creating appropriate responses to given inputs.

Challenges we ran into

It was difficult to fine tune the number of epochs of the machine learning algorithm in a way that it yielded the best final results. Using many of the necessary frameworks and packages generally posed a challenge as well.

Accomplishments that we're proud of

We were impressed by the relative efficacy and stability of the final product, taking into account the fast-paced and time-sensitive nature of the event. We are also proud of the strong bonds that we have formed among team members through our collaborative efforts.

What we learned

We discovered the versatility of machine learning algorithms but also their limitations in terms of accuracy and consistency under unexpected or ambiguous circumstances. We believe, however, that this drawback can be addressed with the usage of a more complex model, allotment of more resources, and a larger supply of training data.

What's next for eTA

We would like to accommodate a wider variety of topics in the program by expanding the scope of the dataset--potentially through the collection of more diverse user inputs from a wider sample population at Berkeley.