Human-in-the-loop LLM content pipeline
SerpAPI keywords → GPT outline → editor approval → article in the same thread to save tokens → two-LLM review → CMS draft.
- Status
- Production
- Period
- Mar 2025 – Feb 2026
- Role
- Data Scientist (AI & Automation), Nam Viet Media
- Code
- Proprietary
- Stack
- OpenAI GPT
- Gemini
- SerpAPI
- n8n
- CMS integration
- Separate LLMs in the review stage
- 2
- SerpAPI keywords
- GPT SEO outline
- Editor approval
- Full article, same thread
- Two-LLM review
- CMS draft
Problem
The content team needed SEO articles at volume without losing editorial control or spending tokens on repeated context.
Approach
- Keywords. SerpAPI finds top-ranking and trending keywords.
- Outline. OpenAI GPT drafts an SEO outline.
- Human in the loop. An editor approves the outline.
- Article. The model writes the full article in the same conversation thread, so the approved outline is not resent, saving tokens.
- Review. Two separate LLMs check each article for accuracy, quality, and SEO.
- Publish as draft. The article is uploaded to the CMS as a draft for final editorial review.
Architecture
A linear pipeline with two human checkpoints (outline approval and final CMS review) and one automated, two-model review stage in between.
Results
Every article passes an editor twice and two independent LLM reviewers once, and the same-thread design avoids resending the approved outline.