100 Days Building PostAll in Public: What I Got Right, What I Got Wrong, What's Next

Short summary

A 100-day retrospective on building PostAll, an AI content generation platform that evolved from a simple OpenAI API script into a multi-CMS publishing tool. Key lessons: generation is the easy part, everything around it (structured output, uniqueness detection, retry logic) is the real challenge. The author's initial Jaccard shingling approach failed to catch paraphrased duplicates, scale assumptions broke at 50 users, and building in public shaped architectural decisions.

• •Generation is the easy part; structured output, uniqueness detection, and reliability are the real challenges
• •Parse-once-render-many architecture saved significant effort when adding CMS integrations
• •Jaccard shingling alone fails to catch paraphrased duplicates; combining with sentence embeddings fixed the blind spot

Generated with AI, which can make mistakes.