Show HN: We Let Machine Learning Design Your Logos

That's usually the first question we get from so many folks. From the surface, it doesn't look like an ML problem but here are some of the places where ML is needed. Given a company:

- What color scheme to use? We generate this using machine learning (random forests)

- What fonts combinations to use (We are trying to do this using font embeddings)

- What icon to select given the company name and description (We use word embeddings right now for this. We are also going to use Conv Nets and Auto Encoders to get better icon representations).

- Out of the hundreds of logos we generate using genetic algorithms, which ones to show to user? (This is an image classification problem and we're working on using Conv Nets here).

These are just some of the aspects where we are applying ML. Some of the features are being used and some are being developed.

Please let me know if you have any other questions. Happy to chat more about this.

From the feedback we've got from different sources including startup communities at reddit, our logos are much better than our competitors (logojoy, brandmark) in terms of diversity and creativity. I would love for you to have a look at the website and try creating some logos.

On the technical side, we are trying to automate as much as we can. One example of this would be a feature that we are working on right now which uses GANs to generate icons from scratch. Another is icon coloring using Conv Nets. These features are still in progress but you can see where we want to go. We want to do away with rules and let the ML algorithms decide everything for us. This addresses one of the limitations of our competitors that we observed - a lot of redundancy because of some rule based aspects of their products.

Happy to chat more about this.