AI Text Humanizer
A private, open-source AI text humanizer: rule-based rewriting in the browser plus an optional deep rewrite by an open model running on the user’s own GPU.
Live deployments
Source repositories
Project overview
Paste an AI-assisted draft and get a version that reads like a person wrote it. The instant mode replaces more than 60 stock phrases, cuts filler and chatbot residue, thins stacked transitions, reduces em dashes and splits overlong sentences, listing every edit. The deep mode loads Qwen 2.5 or Llama 3.2 through WebLLM and rewrites on the device, so the text never leaves the browser.
What I built
- 01
Built an ordered rule set with case-preserving substitution and repairs for verb endings, so rewrites stay grammatical.
- 02
Added an analysis pass that scores AI-writing habits before and after: stock-phrase density, sentence-length variety, em-dash rate and contractions.
- 03
Integrated WebLLM for an on-device rewrite over WebGPU, loaded only on demand and cached by the browser after the first download.
- 04
Ran the rule pass over the model’s output too, catching the stock phrases small models still produce.
Implementation notes
No server is involved in either mode, so drafts stay private by construction.
The pattern score is described as a style measure, not a detector, and the tool makes no claims about passing AI detection.
Technology stack
- TypeScript
- WebLLM
- WebGPU
- Vite
More open-source projects
Need a similar product?
Let’s discuss the workflow, website, or application you want to ship.