01 / 02 · aihumanize.tech product principle
“We wanted the honest version of what other tools show as a fake percentage. So we fine-tuned our own model until the honest version was zero.”
aihumanize.tech
Design principle #2 · 2025
← All case studies + AI/ML · Live · 2025 · aihumanize.tech
In-house product · aihumanize.tech A humanization engine for AI-generated content
The humanization tool that skips the arms race.
One pass. Eight registers. Sentence-level control. Built by people who studied how detectors work — and fine-tuned our own model so you don’t have to think about them again.
API round-trip per humanization
1
Single-pass rewrite — no detect-rewrite-detect loop
Register modes shipped
8
Free · Standard · Academic · Simple · Formal · Informal · Expand · Shorten
Sentence alternatives per click
2–4
Refine any sentence in place
Detected on GPTZero, Winston, Originality
0%
Fine-tuned model tuned around perplexity + burstiness
Year
Live · 2025 · Stripe active
Industry
Content SaaS · writers, students, marketers
Role
Product, design, engineering · in-house at Creative Mantra
Stage
Freemium · Premium at $49.99 /mo
Scope
What we were hired to build.
AIHumanize.tech is a specialised text-refinement engine that rewrites AI-generated content to mimic human writing patterns — restructuring sentence syntax, replacing over-flagged AI vocabulary, injecting natural variation and adjusting tone. It offers eight humanisation modes (Free, Standard, Academic, Simple, Formal, Informal, Expand, Shorten) and sentence-level refinement via alternatives popups.
How we planned
One pass, a purpose-built API.
The architecture routes every humanisation request to a purpose-built API with a minimal payload and returns transformed text; a regex-based, mode-aware fallback engine catches API failures. The frontend enforces word counting (500 words free, 5,000 premium), rate limiting (50 daily free, unlimited pro), and sentence-level alternative extraction. Auth is email/password plus Google OAuth; payments are Stripe; persistent history is Supabase.
How we delivered
A live product with paying customers.
AIHumanize shipped as a live product with a free tier and Premium Monthly ($49.99) and Premium Yearly ($449) plans. Users paste AI-generated text, pick a mode, receive output, and can click any sentence to see two to four alternatives. An admin dashboard tracks payment logs and user statistics via Highcharts. Rate limiting is enforced per user on a rolling 24-hour window from first use — not UTC midnight. The FAQ explicitly disclaims undetectability: a legal-safety statement designed in from day one.
What we learnt
One-pass output was enough.
The humanisation engine is a single-pass rewrite — not a detect-rewrite-detect loop — and users found one-pass output sufficient. Sentence-level alternatives are a secondary UX feature but drive most engagement, because users prefer refining output manually to accepting the whole thing.
How the client improved
Free-to-paid on Stripe.
AIHumanize reached live operating status with paying customers on Stripe. The free-to-paid conversion model works: trial users who hit the 50-per-day limit see an upgrade modal, and revenue flows from Monthly and Yearly subscriptions with proration on plan changes.
02 / 02 · aihumanize.tech engineering principle
“One pass is enough — if the model is right. We built the model, so we skipped the detect-rewrite-detect loop everyone else is running.”
aihumanize.tech
Engineering principle #1 · 2025
Stack · in receipts
The engineering, in one page.
Runtime
Next.js + Supabase
Purpose-built API + regex-based mode-aware fallback
Auth
Email + Google OAuth
Persistent history on Supabase
Billing
Stripe
Premium Monthly $49.99 · Premium Yearly $449
Rate limits
500w / 50/day free
5,000w / unlimited pro · rolling 24-hour window
Perplexity + burstiness tuned in the model. SEO preservation on. Regex fallback catches API failures with mode-aware rewrites. Highcharts on the admin dashboard. Rolling 24-hour rate limit per user — not UTC midnight. The FAQ explicitly disclaims undetectability: a legal-safety statement designed in from day one.
Next case
TTMS Mailer — an autonomous freight agent that never books a load it can’t verify.
Built for Red Dog Logistics. Lives in the dispatcher’s own Outlook. Scores carriers on a hundred-point scale, negotiates within margin guardrails, and stops the moment it isn’t sure. Nine cron jobs, a $0.01 agent cycle, and 455,057 rows in production.
Have something worth building?
You THINK, We BUILD.
Send a paragraph about what you’re building to hello@creative-mantra.com. We reply within 24 hours with next steps and, for qualified scopes, a fee letter within four working days.