AI Content Automation Platform

Project Overview

A self-hosted AI content platform I built solo from 0 to 1. At its core is a reusable content pipeline: n8n schedules topics and generation, an LLM reviews quality, and Chinese and English editions are localized and published separately, wired to GA4 conversion tracking and SEO/GEO structured data. The first live topic is Zi Wei Dou Shu (aiziwei.online). Beyond the blog, it offers a Zhongzhou-school chart calculator, GPT-4o interactive Q&A, and ECPay paid plans, proving the full loop of content driving traffic and tools converting to paid.

Key Achievements

• Automated content pipeline: n8n publishes twice a week across rotating content types (solar terms, monthly analysis, beginner series, life topics), with named citations from international blogs, internal links, and an LLM prose review before publishing
• Bilingual localization: Chinese and English editions are written separately rather than translated, reaching audiences in both languages
• Product loop: free chart and 3 Q&A per month, NT$199 for one hour of unlimited Q&A, with ECPay connecting the free-to-paid conversion path
• Performance and SEO/GEO: cut homepage LCP from 14.9s to 2.4s (84% faster) and reached a performance score of 86; added JSON-LD, hreflang, and llms.txt
• Analytics: built GA4 conversion events (chart, Q&A, begin checkout, purchase) and a real-human traffic view so decisions rest on actual user behavior

Challenges

• Balancing scale and quality in AI content: automation raises output, but readers and ad platforms both expect content with a point of view and sources
• Consistency of the LLM review gate: the same content needs to get a stable verdict on repeated runs
• Complex domain logic (Zhongzhou-school transformations, star system) that had to be translated into programmable rules

Solutions

• Shifted from mass production to curation: focused on evergreen topics, required named sources and a single throughline, and held to 'AI-assisted, human-supervised'
• Traced the gate's instability to its root cause, tuned model parameters and prompt structure, and stabilized the quality gate
• Built a TypeScript chart engine and a Zhongzhou-school knowledge base, paired with layered AI interpretation

Impact

• Delivered an end-to-end automated pipeline from topic selection and generation through review, publishing, and tracking, running continuously
• Designed with the topic as a parameter, so the pipeline can extend to other content domains
• Supports Traditional Chinese and English

Tech Tags