Verza

Inspiration

We saw Verizon had a chatbot of their own for customers, but did not have one for investors. We seized on this opportunity, as chatbots can be a help for a wide variety of the population. Especially investors because it could save them time from having to dig through documentation.

What it does

Investor relations page focused chatbot that uses a Natural Language Processing (NLP) model for financial questions about the company and answers to them.

How we built it

Brainstorming business needs and possible solutions.
Collecting data from Quarterly earnings calls and incorporating this into our model.

Challenges we ran into

The trials and tribulations that came with completing our first hackathon, namely the complexities behind integrating a full stack machine learning application between two people, while learning new technologies.

Accomplishments that we're proud of

Deploying a machine learning application on a website. Coming in this seemed like a foreign idea to us. How would we deploy something that is typically built within a specific python environment such as anaconda to a webserver? But thanks to our ability to quickly learn Tensorflow.js, we were able to integrate a version of the TensorFlow's BART model in JavaScript.

What we learned

TensorFlow can be deployed on websites and mobile apps with Tensorflow.js.
The main components to NLP's are encoders, a neural network architecture (typically recurrent), and a decoder.
The more partners with a wide variety of skills the better. As being two people, we found it would have been beneficial to connect with others. For our next hackathon in order to be able to multiply our impact that will certainly be a goal.

What's next for Verza

Further improving model architecture by applying transfer learning on Keras's chatgpt2 model.
Implementing a more robust user interface through React.
As well as other changes that come along with making the code more sustainable and understandable.