talks.soubiran.dev publishes my conference and meetup slide decks. It also supplies the localized metadata, summaries, transcripts, and statistics used by soubiran.dev. The archive's root redirects readers to the talk index on the personal website.
One workspace per appearance
The public pnpm monorepo keeps a separate Slidev workspace for each event appearance. Each dated directory contains the slides and public assets for that presentation. Workspaces share dependency versions, root commands, and a local Slidev theme.
A talk's src/package.json defines its date, source language, topics, event, location coordinates, and resource links. English and French metadata each have a title, description, and globally unique UUID. The metadata locale is separate from the language spoken on stage, so an English page can describe a talk given in French.
The date and package name define the public deck path. The localized UUIDs identify the content in catalogs and in the personal website's interactions.
Validated metadata and Markdown
The build compiles the Slidev workspaces into a combined dist directory. Generation scripts validate the talk packages and publish:
talks.en.jsonandtalks.fr.json, arrays of localized talk records.meta.en.jsonandmeta.fr.jsoninside each talk's public directory.statistics.json, with totals and counts by year, event, title, recording availability, and city.
Each localized record includes event details, topics, resource links, and artifact filenames. The shared @soubiran/talks package defines the schemas and types used by the archive and the website.
Every talk must have summary.en.md and summary.fr.md. A talk with a recording must also have transcript.en.md and transcript.fr.md. Validation rejects missing required artifacts, transcripts attached to talks without recordings, duplicate locale IDs, and dates that do not match the event directory.
These checks run before the catalogs are written. Consumers can use the declared artifacts to decide which documents to fetch.
Publication on the personal website
soubiran.dev fetches both localized catalogs during its build. It then reads per-talk metadata and imports summaries and transcripts into separate Nuxt Content collections. The generated talk pages display the summary, available transcript, recording, event location, and resource links.
The website reads statistics.json for its statistics section and uses event coordinates for the talk map. Resource clicks and recording interactions are tracked through the website's analytics composable.
Talks with summaries also appear in the website's /pages.en.json and /pages.fr.json catalogs. Their Markdown documents combine event metadata, resources, the summary, and the available transcript.
mcp.soubiran.dev reads those website catalogs and retrieves the resulting Markdown. The MCP does not need to read the archive's talk catalog or select transcript files itself.
Assets and dependent deployments
Thumbnails and localized Markdown artifacts live in each deck's public directory and ship with the Slidev build. PDF and audio exports use dedicated asset hosts. Generated redirects expose resource paths such as /pdf, /audio, /recording, /transcript, and /src. The transcript redirect points readers to the talk page on the personal website.
Cloudflare Builds publishes the combined static output. After a push to main, GitHub Actions sends a request to redeploy.soubiran.dev with the website's deploy hook and the talks Worker as the deployment to wait for. Cloudflare Access service credentials protect that request.
The website rebuild starts after the archive deployment completes, so it imports the published catalogs and artifacts rather than racing their deployment.
Discussions
Loading discussions...
Add a comment
Checking your session...