soubiran.dev is my personal website. It brings posts, series, projects, talks, and infrastructure documentation into one Nuxt 4 application. The source repository contains the content and the local Nuxt modules that implement the site's features.
Static pages, build-time content
nuxt generate produces the website in .output/public. Wrangler deploys that directory to Cloudflare Workers Static Assets, which serves the generated pages without running Nuxt for each visit.
Nuxt Content reads English and French Markdown into collections for pages, posts, and series. The infrastructure documentation lives in content/infra/, stays English-only, and uses the same page collection and shared layout. Its ecosystem and dataflow diagrams use Vue Flow.
The build also imports remote data. The projects section fetches GitHub repository metadata. The talks module reads both localized catalogs from talks.soubiran.dev, fetches each talk's metadata, and imports summaries and available transcripts into separate Nuxt Content collections. Talk pages combine those documents with recordings, slides, resource links, and event locations. The talk index includes a map and aggregate statistics.
Nuxt UI supplies interface components, with Vue and Tailwind CSS for the application code and styles. Local modules organize content, authentication, comments, reactions, feedback, analytics, and the assistant.
A catalog for readers and tools
The website publishes /pages.en.json and /pages.fr.json. Each catalog is an array of records with a stable ID, locale, content type, title, description, published URL, and explicit markdownUrl. Dated content includes a publication date. Translation links connect records available in another locale.
The catalogs include ordinary pages, infrastructure documentation, posts, series, series articles, and talks with summaries. Infrastructure documents use the page type. Consumers follow markdownUrl to retrieve a document rather than infer its location from the HTML URL.
The /raw/ endpoints export readable Markdown. A talk document includes its event metadata, resource links, summary, and transcript when available. Generated documents for index pages include the corresponding content lists. The site also publishes RSS feeds and LLM-oriented content through nuxt-llms.
mcp.soubiran.dev reads both page catalogs. The website therefore publishes the discovery metadata and Markdown used by the MCP tools, including the talks imported from the archive.
Interactions belong to the API
The generated pages do not own mutable user data. api.soubiran.dev handles accounts, comments, reactions, feedback, messages, and notifications. It also receives assistant requests and supplies the data used by the website's administration interface.
Browser requests include session credentials. Writes use Laravel Sanctum's CSRF cookie, and Pinia Colada manages queries and mutations. Shared page components associate content interactions with a stable contentId.
The assistant can include the current page as context and displays streamed responses, tool activity, and usage information. Its conversation requests go to the API rather than the static hosting Worker.
PartyKit handles the live viewer count separately. When a host is configured, the browser connects to the soubiran-dev room and receives the active connection total.
Analytics, errors, and metadata
The shared useAnalytics composable sends page views and interaction events to Umami. Events cover content links, language changes, talk resources, comments, reactions, feedback, assistant activity, and infrastructure diagram controls. Sentry captures errors and associates each release with the build's Git commit.
The site configuration defines https://soubiran.dev as the canonical URL, English and French routes, a sitemap, a Person identity, and generated Open Graph images. Administration pages and authentication callbacks are excluded from indexing. Raw Markdown is excluded from the sitemap.
Discussions
Loading discussions...
Add a comment
Checking your session...