An automated pipeline of AI models writes every article on this site. This page explains each step, the sources the pipeline uses, and what it does not do.
The steps
- Read the repository. The pipeline reads the repository through the GitHub API: its description, star count, main language, file tree, README, and a selection of source files. An AI model summarizes the code and how it is built.
- Research the context. An AI model with web search reads public pages about the project and about similar tools. When it is set up, the pipeline also searches recent public posts on X about the repository.
- Plan the article. An AI model chooses the angle, the main questions, and the outline.
- Write the article. An AI model writes the text from the research notes and the plan.
- Add images and diagrams. An image model makes the illustrations. Another AI model draws diagrams of how the project works.
- Add margin notes. An AI model writes short notes that explain terms and details.
- Check and publish. Automated checks test the structure of the article. If an image, a diagram, or a note fails, the article goes live without it. Then the article is published on this site.
Sources
The main source is the repository itself. Other sources are public web pages and public posts on X that the research steps find. Articles link to the repository on GitHub.
Models
The pipeline uses large language models and image models from several providers through OpenRouter. We change models when our tests show better results.
What we do not do
- No person reads or edits an article before it goes live.
- We do not run the code. An article describes the code as it is written. It does not test it.
- AI models make mistakes, so an article can contain errors. Check important facts in the repository itself.
Dates and updates
Each article shows the date it was first published and the date of its last update. The pipeline can write a new version of an article later, for example when the repository grows.
Feedback and corrections
Readers can vote on each article. We use the votes and comments to find weak articles and to improve the pipeline. To report an error, see the contact page.