Skip to content

How a pipeline runs, end to end

Whether the input is a street photo or a research query, every project follows the same four stages.

Capture & research

A photo lands with EXIF/GPS data, or a research agent pulls candidates from Reddit, Google, YouTube, and web search.

AI generation & fact-check

Claude drafts captions, blog posts, and scripts — grounding and verifying real people and places before anything is written.

Human approval

Nothing publishes automatically. A token-secured approval page lets me review, edit, or retry every job.

Publish & notify

Approved content posts to WordPress and Instagram via their APIs, with an email notification confirming what shipped.

Three projects, side by side

Same philosophy, different inputs and outputs.

Comparison of Shineharder's three automation projects
Project Purpose Primary input AI model Status
Graffiti Pipeline Photo → social + blog publishing Street photo + GPS Claude Sonnet & Haiku Live
Tattoo Discovery Engine Trend research → blog + video scripts Reddit, Google, YouTube, web search Claude Sonnet (research) & Haiku (copy) Live
Health & BJJ Journal AI-built training plans + performance metrics Workout & BJJ training logs Planned Coming soon

Core stack & background

Before Shineharder, my engineering background was Cisco firewall development and network security — the systems-level discipline that now shapes how I design AI pipelines to be resilient, retryable, and safe to run unattended.

PHP 8.4 Anthropic Claude WordPress REST API ImageMagick / EXIF Cisco firewall development OpenStreetMap Nominatim YouTube Data API Cron-driven job queues Git-based deploys

Want to see the projects in action?

Visit the live sites, or get in touch about the work.

Back to projects