Sergey Tselovalnikov built Fontogen, an AI that makes fonts

Sergey Tselovalnikov built fontogen, a machine learning model that designs a new font from a short description. Type a few words like "bold, sans" or "techno, sci-fi, extrabold" and it produces a real TrueType font file. He built it from scratch as a weekend project, trained it on 71,000 fonts, and the repository has more than 500 stars on GitHub.
What he built
Most image models draw pixels. A font is harder: it is a set of vector shapes, one for each letter, and every letter has to look like it belongs with the others. Sergey's write-up, "Hey, Computer, Make Me a Font", covers how he got there.
He started by reproducing a research paper on generating vector images, then realized fonts could be generated the same way. The model reads the description with a pre-trained BERT encoder and then writes out the glyphs one token at a time. The current version produces a capital-only set of letters.
Why it is impressive
He did this alone, on a single GPU, to learn how generative models are built from the ground up. That meant assembling a dataset of 71,000 distinct fonts, summarizing their descriptions into a few keywords each, and solving the hardest part of the problem: keeping every letter consistent in style. The README asks that anyone retraining it use only open source fonts.
The rest of his GitHub shows the same habit of going deep into the stack. jnr-fuse is a FUSE implementation in Java, so Java programs can act as file systems, with more than 400 stars. firewalker is a testing framework for Cloudflare firewall rules, and he pushed to it as recently as August 2026.
His GitHub bio is his whole philosophy: "It is vital to understand the whole stack of technologies you use from top to bottom."
He lives in Sydney, and his day job is staff software engineer at Canva.
What Vamo sees
- devRank tier: Cracked, the top tier Vamo gives.
- Level: senior, from Vamo's read of his work and career.
- 4,267 contributions on GitHub in the year before this post.
See Sergey's profile on Vamo or their GitHub.
Find engineers like this
Vamo ranks engineers by what they have built and shipped, so you can search for the work instead of the title. Add Vamo to your AI tool and ask for people who build like Sergey, or get a demo and we will run the search with you.
Written by Cameron Whiteside. How we pick these engineers. More posts are on Cracked engineers.
Is this you? Email support@vamo.app and we will edit or take down this page.