Play › Font Pairing Switcher
Type Bench, a font pairing proofer
Pick a display face and a body face, then read them on a real page instead of a specimen sheet. Fifty-odd Google Fonts are wired in, and you can drop in a ZIP of your own.
Loading Font Pairing Switcher…
How it got made ✳
An AI-assisted build, with the taste kept human
Structured output, not vibes
The 50-odd pairings are not hand-typed. I had Claude return a strict JSON schema — family, classification, x-height, recommended pairing — then validated it against the Google Fonts API before it ever hit the UI.
Asking for prose would have meant a day of cleanup. Asking for a schema meant a parser and ten minutes.
The model as a critic
Once the bench worked, I fed pairings back to the model as a typographic critic: where does contrast collapse, which pairs fight at small sizes?
It caught real clashes, and also confidently defended bad pairs — useful either way, as long as you treat it as a second opinion rather than a verdict.
Where I overruled it
The ZIP-upload flow was mine, not the model’s. AI is good at the space of things that already exist; the interesting bit was the case it had not seen.
I use AI to widen the option set, then apply the judgement it does not have.
- Claude (Sonnet)
- JSON-schema output
- API validation
- Model-as-critic loop
- Google Fonts API
- Client-side ZIP parsing
Let’s talk! ✳
Work, weekend nonsense, or anything in between.
- amanibnsalim@hotmail.com
- linkedin.com/in/aman-salim1
- Phone
- +91 91676 09846
- Based in
- Bengaluru, India