Blog-to-Carousel Automator

I built a simple pipeline that starts with extracting clean article content from a URL. That cleaned text is passed to an LLM to generate a fixed number of insights, structured as JSON and constrained to fit predefined slide limits. If the output doesn't meet those constraints, the workflow fails early instead of producing something broken.

For visuals, I avoided generating full slides entirely. Instead, I generate text-free illustrations and keep text rendering separate. The final slides are assembled programmatically with a fixed layout—same canvas size, same spacing, same typography every time. Nothing is inferred at this stage, which removes an entire class of errors.

The whole workflow is orchestrated in a self-hosted setup, which gives me control over execution, cost, and iteration speed.

The Impact

The result is a system that generates a publish-ready carousel in under a minute, with a success rate of around 95%.

More importantly, it's maintainable. If I want to change the tone, I adjust the content step. If I want to change the design, I update the rendering template. I'm not relying on a single fragile prompt to do everything.