AI CONTENT PIPELINE
Watch an agent write, rewrite, and ship its own frontmatter.
A real 3-step Groq LLM chain, not a mockup: an honest scaffold that marks where you add real code and numbers instead of faking them, a copy-edit pass that strips AI-tell phrases, and SEO frontmatter generation — with token usage, latency, and a live eval shown for every step.
Why a humanizing pass, and why measure it
First drafts from an LLM tend to lean on the same tells — “in conclusion”, “delve into”, “leverage”, “comprehensive”. The second call is a targeted rewrite instructed to avoid that exact list, and the metrics panel counts how many of those phrases show up before and after — a small, concrete eval instead of just trusting that it worked.
What actually happens on submit
Your content type, topic, keywords, tone, and length go to a Next.js route handler, which makes three sequential calls to Groq’s openai/gpt-oss-20b model — scaffold, copy-edit, then a JSON call for the title, slug, description, category and tags. You choose what it writes: a marketing page, a service offering, a sales page, a product launch, a company overview, an explainer, a case study, an event page or a technical article. The scaffold step is told to leave [TODO: …] placeholders wherever it would otherwise have to invent a price, a client, a result, a date, a benchmark or personal experience — an honest draft you finish, not fake content you publish. Nothing is saved anywhere; the output only exists in your browser for this run.
With “also build a landing page” on (the default), a 4th call writes only the page copy, as validated JSON restating the article. The page itself is rendered from five hand-built designs — editorial, product split, bento, minimal and bold — each with its own type and palette, icons and an illustration that fit the kind of page, the full copy-edited content and a contact section that opens a drafted email. All five come back from one run, responsive down to 300px, so you can switch styles and download the one you want without spending more tokens.
FAQ
AI content pipeline questions
What each step does, what it will not do, and what happens to your key.
6 questions
What does each step do?
Step one writes a first draft of the page type you chose. Step two copy-edits it to remove common AI phrases. Step three produces the title, slug, description, category, tags and reading time as SEO fields. It then lays the result out in five designed page styles.
Will the text pass as human-written or beat AI detectors?
That is not what it tries to do. It leaves [TODO] markers wherever only you know the facts, such as real numbers, code or experience, and it never invents a voice or credentials. Fill the markers with real details before you publish.
How many free runs do I get?
A few per day in your browser, with no signup. After that you can paste your own Groq API key to keep going.
What happens to my API key?
It is used only for that one request. It is never logged or stored.
Which model does it use?
A model hosted on Groq. If Groq is unavailable it falls back automatically to Cerebras, so a run can still finish.
Is the result ready to publish?
Treat it as a scaffold. Read it, replace every [TODO] with something true, and edit it in your own voice first.
Free tool by Adesh Shukla — runs a real 3-step Groq LLM pipeline server-side. A few free runs per day, no signup; after that, paste your own Groq API key (never logged or stored, used only for that request).