SaaS · Featured · 2023
StealthWriter.
Every word, human.
presentsAdmit one
Act I
The challenge
The AI models were the easy part, they lived on someone else’s backend. The real work was everything a paying product needs wrapped around them: metering every word fairly, stopping abuse before it ever reached the models, and keeping Stripe and the quota ledger from ever disagreeing about what a user was owed.
Act II
What I built
The web platform behind stealthwriter.ai. The humanization and detection models run on a separate ML backend; this repo is everything around them. I built the product layer: a word-metered quota engine, per-request anti-abuse, async job orchestration, and Stripe subscriptions kept in sync with all of it.
Intermission, the demo
The product, playing itself
Leveraging cutting-edge algorithms, our platform seamlessly optimizes content workflows to maximize efficiency and drive impactful outcomes.
Our platform quietly rewrites the busywork out of your content, so it reads like you wrote it on a good day.
the pass sweeps once on scroll-in · the fuchsia is StealthWriter’s, not ours
The cast
What actually makes it work
Read the card, then flip it to watch the mechanism run.
0x01
Word-metered quota engine
Two point buckets drained in order (monthly first, then purchased) with split-consumption accounting, an RPC audit ledger, and three refund paths for throttle, concurrency rejection, and ML failure.
how it works
Word-metered quota engine
0x02
Per-request anti-abuse
Every rewrite passes an encrypted device fingerprint and device limit, an IP registry with a 10-trials-per-day free cap, a 5-second throttle, and a suspension flag before it ever reaches the models.
how it works
Per-request anti-abuse
0x03
Stripe to quota sync
Webhook events drive welcome / reset / update RPCs (×12 for annual plans) so a subscription change reconciles the quota ledger without a human in the loop.
how it works