I built CosmicMeta.ai, a tech platform that publishes daily analysis on AI,
machine learning, and emerging tech.
The interesting technical bit: every article goes through a two-pass
humanization step that detects and rewrites 24 specific AI writing patterns
(significance inflation, copula avoidance like "serves as" instead of "is",
em-dash overuse, formulaic conclusions, etc.). It's based on research from
the blader/humanizer framework.
Tech stack: Spring Boot, OpenAI + Perplexity APIs, WordPress for publishing,
Firestore for data. The pipeline is fully automated - from topic selection
to research to writing to humanization to publishing.
I'm curious what HN thinks about the humanization approach. Is stripping AI
patterns enough to make AI-generated content genuinely useful, or is there a
deeper issue with AI-written tech analysis?
The interesting technical bit: every article goes through a two-pass humanization step that detects and rewrites 24 specific AI writing patterns (significance inflation, copula avoidance like "serves as" instead of "is", em-dash overuse, formulaic conclusions, etc.). It's based on research from the blader/humanizer framework.
Tech stack: Spring Boot, OpenAI + Perplexity APIs, WordPress for publishing, Firestore for data. The pipeline is fully automated - from topic selection to research to writing to humanization to publishing.
I'm curious what HN thinks about the humanization approach. Is stripping AI patterns enough to make AI-generated content genuinely useful, or is there a deeper issue with AI-written tech analysis?
https://cosmicmeta.ai