Humanizer

Inspiration

Everyone drafts with AI now, but the raw output reads flat and gets flagged by detectors. I kept rewriting AI drafts by hand to make them sound like a person, and realized that was a product.

What it does

Humanizer takes AI generated text and rewrites it so it reads naturally and passes AI detectors. You paste a draft, pick one of eight writing styles, and get a version that keeps the meaning while dropping the robotic patterns. It supports over 100 languages and keeps citations and formatting intact.

How I built it

The app is an Astro and React frontend on a self hosted Postgres stack. Under the hood it runs a multi pass rewriting pipeline on top of modern language models with custom post processing that removes the statistical fingerprints detectors look for.

Challenges I ran into

The balance is delicate. Rewrite too little and it still gets flagged, rewrite too much and you lose the original meaning or introduce errors. Tuning the passes so the output stays faithful while reading human took a lot of iteration and evaluation.

Accomplishments that I'm proud of

It processes text in seconds, works across many languages, and reliably passes the popular detectors while still saying exactly what the writer meant. We never store content longer than needed to produce the rewrite.

What I learned

Writers do not want a black box, they want control over tone and confidence that the meaning survives. Giving them styles and keeping formatting intact mattered more than raw detector scores.