Content Automation Pipeline
Idea in, fact-checked post and on-brand carousel out. I built it to run my own LinkedIn presence, and it has been running daily ever since.
Engagement
Built for myself — running daily
2h → 15 min
per day of content production
The problem
What this replaced
The manual process, before
- 01Brainstorm a topic.
- 02Search for facts and verify them.
- 03Draft the post.
- 04Design carousel images by hand, on-brand, every time.
Who has this problem
Engineers and creators maintaining a technical brand through consistent publishing.
Why it costs money
Consistency drives reach, but one to two hours a day of manual production comes straight out of billable work.
The build
What I built
Takes a raw idea, fact-checks it with live web search, drafts the post copy, then generates HTML against a set of brand guidelines and renders it to an image with Playwright — so every visual is consistent in color, logo placement, typography and layout without a designer in the loop.
Automate personal content production end-to-end, from raw idea to publish-ready post and visuals.
Capabilities
- Fact-checking via live web search
- Post copy drafting — hook, body, CTA
- Brand-guideline-driven HTML generation
- Playwright HTML-to-image rendering
- Generalizes to any brand system by swapping the guidelines
How it runs
The workflow
Trigger
A raw idea or topic prompt, plus the brand guidelines.
- 01Capture the topic and viewpoint
- 02Run a live web search to verify facts and pull fresh data
- 03Draft the post copy — hook, body, CTA
- 04Generate HTML for the carousel, following the brand guidelines
- 05Render the HTML to an image with Playwright
Output
A finished post draft plus rendered carousel images, ready to publish.
Architecture
How it fits together
Components
- Idea intake
- Content agent — running on Claude Code
- Live web search (Tavily)
- Brand guideline set
- HTML generator → Playwright renderer
Data flow
01
Idea → fact-check via search
02
Draft copy
03
Generate brand-conformant HTML
04
Playwright render → image files
The guardrail
What it is not allowed to do
Search-grounded, human-published.
Drafting is grounded in live search results rather than model recall, which is what keeps factual errors out of published technical content. The final publish step stays manual on purpose — nothing goes out under my name that I have not read.
Stack
What it is built on
- Runtime
- Claude Code
- Search
- Tavily
- Rendering
- HTML generationPlaywright
- Design system
- Brand guidelines — color, logo, typography, layout
Hard parts
What made it difficult
- 01
Generating layouts that stay on-brand across wildly different content lengths.
- 02
Grounding claims well enough that technical readers do not catch errors.
- 03
Keeping visual consistency without hand-designing each asset.
Outcome
What changed
2h → 15 min
daily production time
Content production time fell from one to two hours a day to under fifteen minutes.
Search-grounded drafting reduced factual errors in published technical posts.
The same pipeline generalizes to company content — swap in a company's brand guidelines and it produces on-brand visuals for their pipeline instead of mine.
Limited availability
Tell me the problem.
I’ll tell you if an agent is the answer.
A 30-minute call, no pitch. Describe the manual loop that hurts and I’ll give you a straight read on whether this is worth building — including when the honest answer is a script, not an agent.
30 minutes · video call · no pitch