-
Login Page
-
Sign Up Page
-
Ai Assistant
-
Home Page
Inspiration
What inspired us was the Morgan & Morgan challenge in creating a Virtual Assistant using AI. We wanted to take this opportunity to learn the ways of Neural Networks and build an AI from scratch. We also wanted to expand our knowledge in JavaScript and learn how the back-end and front-end connect.
What it does
It's a functional website that allows you to chat with a Virtual Assistant. You can feed it legal case questions and it will extract the vital information, throw this into the AI component, and attempt to give you the best reply possible. It also has login and sign-up pages in case the user wants to save their data. To the left of the website, there is a search bar that could allow you to find previous conversations with Morganai.
How we built it
We divided into two groups: one responsible for creating and training a Neural Network from scratch, and the other for developing a functional UI that connects the AI to the ChatBody. For the neural network, we wanted it to recognize keywords that the user types into the chat box and output a specific response to it. To build the Neural Network we started by researching YouTube videos that explained the theory and math behind it. We then created the model from scratch using the knowledge we gained from those videos, setting up the neurons, making layers of them, connecting them by weights, and training them. This resulted in a basic, 3 layer neural network. One layer of input nodes, one layer of hidden nodes, and one layer of output nodes. To build the ChatBody assistant we first started by creating a basic API using JavaScript and making sure the basic connections were working. We then elaborated more on the home page using CSS for styling and created a header with functional links and a chatbox.
Challenges we ran into
Working with Git/Github proved to be an initial challenge, as few of us either were rusty or did not know how to utilize it in a team-based environment. It would lead to some complications down the line that we had to power through. Understanding the math behind the neural network was incredibly difficult, leading to some approximations in the backpropagation calculations that, while initially seemed to work, did cause the network to consistently misbehave when introduced to a larger and more complicated dataset.
Accomplishments that we're proud of
Stepping outside our comfort zone to attempt to create our own AI from the knowledge we gathered instead of just going with a pre-trained AI model.
What we learned
We about the process of creating a basic neural network, and how the math required for it is complex, and something that cannot be so easily side stepped. We learned how to use React and CSS together to make different pages and how to link them using routers. We also explored the different uses of branches in GitHub and the importance of effective git management.
What's next for Morganai
With more training, Morganai could deliver more accurate and impactful summaries and showcase the saved data on the left sidebar of the website.
Log in or sign up for Devpost to join the conversation.