Namy.ai

Hello! I'm Hugo 👋

I'm always starting projects. That's where the idea for this website came from. As soon as I think of a new project, I start daydreaming about it's name, a website, etc... But most domain names are taken! Or at least it always seems to me that way. Good domain names fly.

So why not, I thought I'd take a shot at it and try to make a solution myself.

So here is Namy: an AI domain generator that's actually good. (hopefully you agree 🤞)

Read more at https://namy.ai/about

# How namy.ai works

If you're reading the about section of a website I'd figure you'll be interested in knowing how it works. So here's a brief overview:

I scrape a large number of websites (~10 million) for their domain name + their contents and some other high level features.

From here on there's lots of cleaning and data processing to create a good dataset (probably the part that took me the longest).

I then train a reasonably big PyTorch model on this dataset, with the aim of generating new, never before seen domains. The model is mostly composed of LSTM layers, and training takes roughly 20 hours on a GPU.
EDIT: Instead of LSTM layers, it's now a customized transformer architecture. Works much better! (and training takes much longer)

After training comes the fun part: generating new domain names! Generation is slightly complex, as there are a few parameters I tune and rules I add, but in essence it's asking the model to predict new sequences.

These AI generated domains are checked for availability using the namecheap API (and a few others if I hit the quota limits).

Finally, they're temporarily stored in a database (PostgreSQL), from which the Bootstrap frontend pulls every ~1 second, until the results are ready and can be displayed.