Both of us have experience working in the AI research field. More times than not we find ourselves scrambling to try and figure out which architecture is the best for our needs. We came up with this idea as a solution.
What it does
This app allows for users to create different problems, add their own dataset, and allow for others to contribute to their initial problem with their architectures. We then rank the architectures on a number of factors such as efficiency, time to complete, and size.
How We built it
We used electron.js to create a nice desktop application for MacOS, Linux, and Windows. The app was organized using React and react-router, to enable code reusability and multi-page routing. The app used various Node.JS APIs to query system information and to run the ML models that the user specified. We then displayed the results of those ML models using another React terminal component.
Challenges We ran into
Lots. First, a general issue in AI is that packages aren't well maintained and often different functions are deprecated without enough notice. Console services in node.js do not cooperate well with command line arguments that create files.
Accomplishments that we're proud of
The idea. The design. And the current functionality.
What we learned
We learned that there is a need for better AI package management.
What's next for Model Bench
We do plan to continue building it in the future as it has real world applications and may actually be used in the future.