AI Credit

Home Screen
Algorithm Details
Team Members
User Input

Inspiration

Machine learning has been applied to many industries, including finance. Specifically, AI has been implemented in fields like investment banking and risk management. We wanted to apply these AI concepts to a credit analysis application using online datasets. However, after speaking with several professionals in the Finance and AI Industries, we discovered that using pre-built AI libraries such as Keras and Tensorflow prevented specialists to customize their AI model to the maximum. In order to really differentiate AI Credit, we decided to completely build our own machine learning library (As shown in the video demos).

Creating Our Project

We built this project in two parts: the back end and the front end. The back end was written in Python and is made up of an artificial neural network (machine learning algorithm) that was built from scratch,which detected patterns in an online dataset. Traditional machine learning libraries, like TensorFlow and Keras, were not used in his implementation. Instead, only the standard computing library Numpy was used. The backend used Django to manage the Http Requests in order to send and receive information from the HTML. The front end is the website written by @Ishaan Ghosh, @22pilarskil, and @Aryan Dugar. It uses HTML and Javascript-- HTML to build the structure and Javascript to implement the trained algorithm obtained from backend. Finally we used to Digital Ocean to host our project.

Challenges

We had several challenges. One problem was connecting Blockstack to our website. We realized too late that Github was static, and to incorporate Blockstack in our website itself, we needed a dynamic website. Our initial idea was changing an HTML paragraph element to detect if a user signed in or not on Blockstack; however, this only worked on the computer hosting the Blockstack server.

Accomplishments

We created a successful neural network built almost completely from the ground. This neural network has a test accuracy rate of 90.2% when evaluated on our database. Incorporating this network into our website was our biggest accomplishment. A video explanation of our neural network is found at https://youtu.be/qcTgUsDvKnY.

Built With

Submitted to

hackMHS VI
- Winner Honorable Mention
Blockstack - Can't Be Evil Contest: Part 2

Created by

I worked on displaying the data sent from the backend, I also was in charge of getting user information to the machine learning algorithm

Liam Pilarski
Hi! I’m Liam, and I am a first year CS major at Georgia Tech. I’m super passionate about AI, and hope to continue using it in cool projects
I worked on the Backend development for AI Credit. Specifically I used Django to handle Http Requests, and I used a self built database to store and transfer crucial data.

Aryan Dugar
Hi my name is Aryan Dugar and I am a student at Millburn High School. I am interested in backend-end development and the world of finance.
Worked on back-end by creating a neural network from scratch (using only Numpy). Achieved 90.2% test accuracy on credit approval dataset.

Ryan Park
A high school student currently studying machine learning.
Ishaan Ghosh