A visual tool for creating machine learning models.

Kogu is a tool that helps to develop AI models easier, faster and with more transparency. It integrates seamlessly with your existing workflow.

Kogu started as an internal project for a data science team. We experienced situations where we wanted to share results, but found ourselves digging in a mountain of unorganized scripts. The concept evolved from a desire to cleanly and concisely store versions, visualize and interprate our ml experiments.

To suit our target audience we wanted to keep things simple. We ordered the branding which we modified to deliver a minimal aesthetic. Animated the logotype and used a very subtle, unintrusive colorscheme.

I wanted to present my findings to my team in multiple remote locations. So I shared screen recordings of my experiments where I could explain the process while presenting the prototype in action.

The tool interpreates values from the scripts and generates a UI from the inputs and outputs. The output creates a real-time log and visualization of the results. The input enables a simple way to tweak and re-run with different hyperparameters.

The project was not completed in its full form due to budgeting constraints. But we got the initial MVP released, that included some basic scaffolding with versioning and logging without visualization an re-running. The project is now open-sourced on github.