VisualML | Devpost

Inspiration

A large part of the team is interning at a medical research and institute and that is where we recognized a problem. Lots of these scientists who are working on cutting edge research have to waste crucial time learning programming, and writing boring redundant boilerplate code if they want to design custom ML models. We realised that there is no reason that training machine learning models has to be this hard. Machine learning has limitless potential and we decided to make it accessible for everyone regardless of prior technical experience. Now scientists, researchers, and even experienced software devs can focus on what they do best, and leave all of the boring, redundant boilerplate and overheads to us.

What it does

VisionML is a simplified, GUI based, user-centred platform that allows you to make fully custom machine learning models to suit your specific needs. You can define your own architecture, choose your own layers, and set your parameters to have a model that works for you. All without ever writing a line of code.

How we built it

VisualML is built using Svelte Kit, tailwind, flask and python. The GUI selections are interpreted to PyTorch and the model can then be trained on our own flask server, your personal GPU onsite, or on a cloud environment like Google Collab.

Challenges we ran into

We started with a lot of good ideas initially. It was difficult narrowing down the ideas into VisualML. We also had difficulty finding proper libraries and frameworks, we found interactJS and Google Cloud, both of which required paid tiers to work with our technology.

Accomplishments that we're proud of

Making an intuitive, simple, and aesthetically appealing, accessible platform for training custom ML models is something that we are very proud of. Now even your grandma can make her own ML models.

What we learned

We learned a lot about low level machine learning architecture and how it all works under the hood. We also learned a lot about cloud training environments, and using Svelte, and flask servers.

What's next for VisualML

We hope to expand VisualML to be monetized, using a B2B SAAS model, so that there can be a free licence to community users, and a development and creator subscription based paid tier. This way VisualML can be profitable yet also be accessible to individuals and hobbyists.