Building a docs-as-code platform
This case study describes how I built a docs-as-code platform for a new AI product team.
Context
I was the only technical writer on an early-stage AI product team. The team had no formal documentation and no publishing workflow. Product knowledge was spread across engineer notes, conversations, and scattered Confluence pages. There was no information architecture, review process, or path from draft to published content.
Constraints
The project had these constraints:
- As the only writer, I had to maintain the system and create the initial content.
- The team had no budget approval for proprietary authoring or publishing tools such as Oxygen XML or Adobe Acrobat.
- The platform had to use open-source, reproducible tooling that the team could run, extend, and maintain without licensing dependencies.
- The architecture had to support reuse of the same setup for other product documentation later.
What I built
I designed and implemented a Git-based documentation system. It included the following components:
- Markdown and MDX authoring
- Pull-request review
- Automated publishing with GitHub Actions
- Product, developer, and API documentation in one site
- Vale style-guide rules for the editor and for pull requests
- Internal hosting, with external publishing planned for later
I chose Docusaurus as the static site generator because it supports sidebar configuration, MDX extensibility, and API reference generation in one build.
How the workflow worked
- The writer updated a file and opened a pull request.
- Reviewers checked the change for technical accuracy and editorial quality.
- GitHub Actions built and published the site after merge.
Navigation, sidebars, templates, and content grouping were defined in version-controlled configuration, so structural changes went through the same review process as content.
I converted the company style guide into Vale rules. Contributors then received style feedback in their editor and automated checks during pull requests.
Outcomes
The project produced these results:
- The team moved from ad hoc notes to a single published documentation site.
- GitHub Actions published new pages within minutes after merge.
- Engineers contributed through the same Git workflows they use for code.
- The team later reused the same setup for other product documentation and for external developer publishing without rewriting content.
What this demonstrates
This project shows my ability to build documentation infrastructure and publishing workflows when none exists yet.