Three passes, so the rewrite never sees the original.
Humanize rewrites AI-generated text into prose a person would write: a three-stage LLM pipeline that first reduces the input to notes, drafts from those notes in a chosen tone, then polishes token-by-token over a stream. No accounts, no login.
- 3
- rewrite stages
- 5
- tones
- 20k
- character cap per run
- 0
- accounts needed
The brief
AI text is recognisable by its phrasing, and a single-pass rewrite keeps that phrasing because the model is looking at it. The fix is structural: make the rewriter work from notes, so the original wording never reaches it.
What shipped
Every line below is in the build, not on a roadmap.
- Stage one reduces the input to flat bullet notes at low temperature
- Stage two drafts from the notes under one of five tone personas
- Stage three polishes, streamed token by token over Server-Sent Events
- A deterministic post-pass strips em-dashes, semicolons and known AI sentence starters, and matches the source paragraph count
- Five tones with their own length ratios; a 20,000-character cap
- Single-file FastAPI backend on Fly.io, scale-to-zero; React front end on Vercel
Want one of these?
The same stack and the same process, priced from the floors on the service page. Every number is negotiable.
Crafted in the dark. Shipped to the world.
Tell us what you are building. You get a private project space with a proposal and a line-by-line quote within a day.