Content Automation System

The Bottleneck: Admin vs. Education

Consistency is the biggest hurdle for technical creators. To publish 3+ tutorials a week, a person would spend 3-4 hours per video on tasks that had nothing to do with teaching:

• Validation: Spending hours researching if a topic was actually trending.
• Context Loss: Manually drafting scripts that often lacked a consistent "brand voice."
• Operational Friction: Manually creating Google Drive folders, Docs, and updating tracking sheets for every idea.

The "Tutorial Factory" was built to solve this by moving the administrative burden to an agentic AI system, allowing me to focus entirely on the final 20%—creative delivery and personality.

The Solution: Agentic RAG Architecture

I built a multi-stage workflow in n8n that runs like an automated research and writing department.

1. Intelligent Idea Validation (The "Researcher")

Using the Perplexity AI node, the system does real-time internet research on trending topics. It's not guessing, it's validating difficulty levels and actual viewer pain points before anything gets made.

2. Brand Voice Injection (The "RAG" Layer)

To keep the AI from sounding generic, I integrated a Pinecone Vector Database. The system pulls my past script snippets and metaphors (embeddings) and feeds them straight into the Claude prompt. That's Retrieval-Augmented Generation doing what it's supposed to do.

3. Automated Production Ops

Once I approve an idea in Google Sheets, the system handles the rest:

• Generates a full technical script with hooks and step-by-step walkthroughs.
• Creates a dedicated Google Drive folder and formatted Google Doc.
• Builds a production "Shot List" and SEO metadata automatically.

The Engineering Choice

• n8n (Self-Hosted): Orchestrates 25+ nodes including complex loops and JavaScript data transformations. Running on Railway keeps costs at zero.
• Claude LLM: Built for higher-level reasoning on technical scripts and strict JSON formatting when I need clean database logs.
• Pinecone: This is the Brand Brain. It lets the system scale without losing the specifics of how I teach.

Governance & Reliability

Automation at scale needs guardrails. This system includes:

• Human-in-the-Loop: AI generates options, but I am the "Gatekeeper." No script is drafted until I manually change a status cell in Google Sheets.
• Self-Healing Logic: Built-in 3x retry nodes handle API rate limits and network timeouts automatically.
• Error Logging: A dedicated "Audit Log" captures every failure, so I can troubleshoot instantly.

The Results

This project isn't just about making videos; it's a blueprint for Learning Operations — the same pattern applies to any team drowning in content admin.

• Scale: Tripled production volume (1 to 3+ videos per week) while reducing total work hours.
• Efficiency: Cut the "Idea-to-Script" lifecycle by 75%.
• Consistency: The Pinecone memory layer preserves brand voice at scale — the system drafts like I write.
• Reliability: Self-healing retries and a dedicated audit log keep it running without babysitting.