Case study · AI agents & automation

A multi-agent content engine that runs itself

We built an end-to-end content pipeline that scouts new stories across the web every day, scores them automatically, and drafts a finished piece - with a person approving only what actually clears the bar.

Talk to us about an AI pipeline See how it works

Domain · Content & media automation Data · Daily multi-source web discovery Output · Scored, drafted content with human sign-off

The challenge

Scouting good stories doesn't scale by hand

Staying on top of what's actually worth covering means checking multiple sources every day - product launches, forum discussions, community threads - judging what's relevant, writing it up, and producing it. Done manually, that's a full-time job before anyone's even started on quality.

The goal was to automate the entire loop except the one judgment call that still needs a person: deciding whether a scored, drafted piece is actually worth shipping.

How it works

From daily discovery to a finished draft

Four stages take a pipeline from "what's out there today" to a piece ready for a human to greenlight.

Discovery scraper Daily scan

An automated scraper checks product-launch boards, tech-news aggregators, forum communities, and RSS feeds every day, surfacing new material before a person would ever see it.

Scoring & filtering Rule-based

A rules engine scores every item 0-100 and automatically flags affiliate-program mentions, so only genuinely relevant material moves forward - no manual triage queue.

Script & voice generation AI-drafted

Items that clear the score threshold get a generated script and voiceover - a finished draft ready for production, not just a summary someone still has to rewrite.

Human-in-the-loop & delivery Approval gate

Nothing gets produced without a person signing off - the score only triggers a review, not an automatic publish. Approved pieces move to production, and a weekly digest goes out automatically.

Why it holds up

Built to run daily without babysitting

What it delivers

Daily coverage across multiple sources without manual monitoring.

Consistent, rule-based scoring instead of ad hoc judgment calls.

Nothing goes to production without a human approving it.

A standing weekly digest with zero manual compilation.

Engineering behind it

Python core orchestrating the full pipeline end to end.

n8n workflows running the automation layer.

Airtable and SQLite tracking state across every run.

Generation via the Claude API, not fixed templates.

Get started

Drowning in sources you can't monitor by hand?

Discovery, scoring, drafting, and a human checkpoint before anything ships - we can build the same shape around your own sources and outputs.

Book a discovery call

Internal Waveo system, described by architecture and outcome.