Inspiration

Thanks for checking out team LonelyStar's project! The inspiration for the project came from the desire to build a project that incorporates machine learning. Out of all the challenges in HACC, the chatbot challenge is one that stood out where machine learning can clearly be used, by understanding user intent through deep learning and transformers.

What it does

The project is called OHABot and it is a chatbot that allows users to type in their OHA-related questions. The chatbot will then try to understand the user's intent and direct them to relevant information.

How we built it

The project was built with React, Django, Pytorch, and HuggingFace. The fronted is rendered by React which sends the user's questions to a Django API. Django then uses the trained model to infer the user's intentions and returns a category that the frontend renders. The deep learning model is a Distilled Roberta transformer model from HuggingFace which I repurposed to support the needs of the app.

Challenges we ran into

There were lots of challenges in this project. The first challenge was ensuring that the frontend can talk to the Django backend. The second challenge was making sure that Dockerfile works properly because Github has a size limit on repos so the model had to be trained inside the image and published to DockerHub. The hardest challenge by far was deployment because free services like Heroku have limits on how much memory the server can use and since this project uses a machine learning model, it was too resource-intensive for trial runs in some platforms like Heroku.

Accomplishments that we're proud of

I am really proud of how well the model understands the user's request. The model was trained with only 100 something lines of data and during my tests, it performed surprisingly well. The model can very easily be improved by simply adding more data to the CSV file and retraining the model. I am also proud of how much the team accomplished because there is only one member on the team.

What we learned

I learned an absolute ton for this project. I learned how to make API calls with react as well as how to integrate React into Django. I also learned a lot about how Docker works from the many failed attempts I had at creating a working image and I learned a little bit about deployment of applications.

What's next for LonelyStar

This project was what I would call a high code solution. Everything from the frontend, backend and deep learning model was built or trained by hand so the process took very long. Currently, the project can be run locally but I will be working on deployment of the project asap. Once deployment of the project is done, I would like to work on the project some more to add a feature that allows people from OHA to add their own categories and templates to the deep learning model so that it can be extended to other questions in the future.

Built With

• django
• huggingface
• javascript
• python
• pytorch
• react