Documentation Systems#

AI-SAI-Generated Summary H-RHuman Reviewed H-EHuman Edited

At Transfer Online, I researched and built a docs-as-code system from scratch using reStructuredText, Markdown, Sphinx, Git, CI/CD, HTML, and LaTeX. After previously working with Microsoft Word, I recognized that mature, text-based formats could support maintainable manuals immediately and more reliable automation later.

Note

Implementation timeline I designed and implemented the underlying documentation architecture, standards, reusable content patterns, and initial publishing automation before generative AI was introduced. Later additions included AI-assisted drafting and maintenance.

View the complete implementation timeline.

With Sphinx and RST, I created customized manuals with reusable admonitions, styles, shared content, and role-specific outputs. The initial manuals and authoring standards were researched and developed manually. Once those patterns were stable, the same source could generate accessible web documentation and LaTeX-based PDFs through CI/CD whenever content changed.

Later, I used Python, model APIs, and agentic workflows to accelerate drafting and maintenance while keeping the established structure and review process. Reusable rules and skills gave the agents explicit style guidance and helped them apply the same content patterns consistently.

PDF pagination remained a difficult part of the workflow. Drawing on my computational-modeling background and linear-algebra coursework, I developed a scoring approach that evaluated candidate formatting changes before publication. The workflow used that result to improve the first PDF build, bringing the output much closer to the expected layout while retaining human review.

The result was a repeatable system for producing and maintaining detailed manuals in synchronized online and offline formats at substantially greater speed.