Inspiration

We wanted to do something silly with 'Anything is a hammer in mind'. We decided to overcomplicate something using AI and Neural Networks for basic four operations was our focus

What it does

A calculator app that creates an AST tree of operations for user to see and uses a pre-trained (by us) Neural Networks to mimic basic four operations. Only has around 1% error rate. The workings of NNs is visualized as well for better clarity and more info.

How we built it

Use TS and Flask for the frontend to create a basic calculator app, visualize NN and ASTs. For training the NNs torch and sklearn was used. The data for supervised learning is very easily obtained as we can create as many basic binary equalities (Ex. (1+1=2)) as we want. The backend also creates an AST to preserve the order of mathematical operations.

Challenges we ran into

Deciding on a project was the biggest challenge. Then we needed to learn how to train sufficiently complicated NN that can emulate addition and multiplication. Thios took alot of training.

Accomplishments that we're proud of

We are proud of the product in general.

What we learned

We learned to prioritize and divide the workload according to our strengths.

What's next for Overcomplicated Calculator

Hosting it on cloud would be the next step

Share this project:

Updates