AI + Data for Science (AIDa4Sci)

Quarto website for the Stanford seminar series AI + Data for Science (CS 292 / EE 292R / PSYCH 292R / STATS 282), Wednesdays 4:30–5:30 pm in CoDA E160. A talk held in another room carries a location field in the quarter's YAML file; see "A talk in a different room" below.

Local preview and build

quarto preview   # live-reloading local server
quarto render    # builds the site into _site/

Publishing

Pushing to main builds and deploys the site via GitHub Actions (.github/workflows/publish.yml). Nothing is committed to a gh-pages branch — the built site is uploaded straight to Pages.

Nothing here needs setting up again. Settings → Pages already has Source set to GitHub Actions, and both the repository and the Pages site are public:

gh api repos/stanford-developers/aida4sci --jq '.visibility'
gh api repos/stanford-developers/aida4sci/pages --jq '.build_type, .public'

Keep it that way. While this repository was private inside an enterprise organization, Pages defaulted to private visibility — served from a randomized *.pages.github.io hostname, behind a GitHub login with organization membership. Returning to that would put the schedule out of reach of every external speaker, and would take the custom domain with it, since a private Pages site cannot carry one.

The workflow also runs on a schedule, Thursdays at 08:00 UTC. That is deliberate: the home page's "Up next" card is resolved against the build date, so without a periodic rebuild the site would keep advertising a seminar that already happened. The weekly run rolls the card over the morning after each Wednesday seminar. You can also trigger a rebuild by hand from the Actions tab.

The stanford.edu address

The site is served at https://aida4sci.stanford.edu, with HTTPS enforced under a GitHub-issued certificate. website.site-url in _quarto.yml matches that address — it feeds the sitemap and the link previews search engines and chat apps show — and GitHub redirects the older https://stanford-developers.github.io/aida4sci/ URL to it.

Two things hold the address up, and neither is visible in this repository:

• a NetDB DNS record, aida4sci.stanford.edu. IN CNAME stanford-developers.github.io.
• the custom domain set under Settings → Pages, stored on GitHub's side.

There is no CNAME file at the repository root. If one is ever added it must also be listed under project.resources in _quarto.yml, or quarto render will leave it out of _site/ and the deploy can drop the domain.

Check both from the command line:

dig +short aida4sci.stanford.edu CNAME
gh api repos/stanford-developers/aida4sci/pages --jq '.cname, .https_enforced'

If the name ever has to be re-established, ask NetDB for a CNAME record, not a redirect. The first attempt produced a host pointed at Stanford's link service (stanford.dns.bl.ink), which answered with a 307 to the github.io URL. That is a forwarder, not a custom domain: the address bar still showed the github.io URL, GitHub could not issue a certificate for the name, and setting the domain under it would have failed verification and unpublished the site. If NetDB will not put a CNAME at that name, the fallback is A records to GitHub's Pages addresses — 185.199.108.153, 185.199.109.153, 185.199.110.153, 185.199.111.153 — but the CNAME is preferred, since it survives GitHub renumbering those.

The /pickdate short link

https://aida4sci.stanford.edu/pickdate sends speakers to the quarter's Microsoft Bookings page. GitHub Pages has no server-side redirects, so pickdate/index.html is a plain HTML page that forwards in the browser (an immediate meta refresh, a script, and a fallback link). It is copied into _site/ only because pickdate/ is listed under project.resources in _quarto.yml; drop that line and the link breaks.

When a new quarter gets a new Bookings page, replace the URL in pickdate/index.html. It appears four times; change them all.

"Up next" on the home page

The home page shows the next seminar automatically — there is nothing to update by hand. ejs/upnext.ejs reads the same quarter YAML files as the schedule (all of them are listed under contents in index.qmd) and picks the nearest date that has not passed. If that date is still unbooked it says the speaker is not announced yet rather than skipping ahead; once every listed date is past, the block disappears rather than showing something stale. Because every quarter of the year is listed, the card rolls from the last Fall talk to the first Winter one on its own.

Adding a speaker

1. Open the quarter's data file: speakers/fall-2026.yml, speakers/winter-2027.yml, or speakers/spring-2027.yml.
2. Replace the tba: true entry for the chosen date with a filled-in block — copy the example in speakers/_template.yml. Fields you omit (url, photo, bio, …) are simply not shown.
3. Drop a square-ish photo into images/speakers/ (e.g. images/speakers/jane-doe.jpg) and reference it in the photo field. If there is no photo, omit the field and a neutral placeholder is used.
4. quarto render.

A talk in a different room

The usual room, CoDA E160, is not written into the YAML. Add location: "Packard 101" to an entry only when that talk meets somewhere else. Both templates then flag it: the schedule card and the home page's "Up next" block get an ember notice naming both rooms ("Packard 101, not CoDA E160"), while every other card quietly shows CoDA E160 beside its date. Leaving location off is what means "the usual room" — do not add it to every entry.

The room links to the campus map by its building name; add location_url: to point somewhere else. The usual room itself is a constant, DEFAULT_ROOM, near the top of both ejs/speaker.ejs and ejs/upnext.ejs; if the seminar ever moves for good, change it in both, and in the prose on the home, schedule, and subscribe pages.

The schedule page also carries a callout listing the exceptions for the quarter, and the home page's Logistics section repeats it. Those two are prose, so update them by hand when the exceptions change.

Quarters and the academic year

The schedule page shows every quarter of the current academic year at once — Fall, Winter, and Spring, each under its own heading — so speakers can see the dates still open for the year, and /pickdate can book any of them. All three quarters have to be on the one page: the calendar feed and the "Up next" card deep-link to a talk as schedule.html#<ISO date>, and a per-quarter page would break those links.

Everything flows from the YAML files, so the whole year is wired up when:

• speakers/<season>-<year>.yml exists for each quarter (e.g. speakers/winter-2027.yml), with one entry per seminar Wednesday — tba: true until a speaker confirms;
• schedule.qmd has a listing (and a heading plus ::: {#id} block) for each file, and a chip for it in the quarter-nav jump bar;
• index.qmd lists each file under the upnext listing's contents.

The calendar feed needs nothing: tools/gen_ics.py reads every file in speakers/, so unbooked dates publish as "speaker TBA" events that update in place when a name goes in.

Adding the next academic year

1. Check the dates against the Stanford academic calendar — drop any Wednesday that is a holiday or recess, as November 25, 2026 was dropped for Thanksgiving.
2. Create the three YAML files and add them to schedule.qmd and index.qmd as above. Update the year in the schedule page's title and the home page's hero eyebrow.
3. Set the Bookings page's availability to match, one window per quarter with the breaks between them not bookable.

Archiving a finished quarter

Once a quarter's last talk has happened (for Fall 2026, after December 2):

1. Create past/<season>-<year>.qmd (e.g. past/fall-2026.qmd) with a title, the plate header-includes, and the same listing front matter as schedule.qmd but pointing only at that quarter's YAML file.
2. Link it from past.qmd, replacing the "no past schedules yet" text the first time.
3. Remove the quarter from schedule.qmd: its listing, its heading and ::: {#id} block, its chip in the jump bar, and any callout that was specific to it (the Packard 101 notice is a Fall 2026 one).
4. Remove its YAML file from the upnext contents in index.qmd, and drop any quarter-specific prose from the Logistics section there.

Leave the YAML file where it is — the archive page and the calendar feed keep reading it. A deep link such as schedule.html#2026-09-23 from an old email will no longer land on the card, but the talk stays browsable in the archive.

Structure

• _quarto.yml — site config (navbar, footer, theme)
• styles.scss — Stanford-cardinal styling, talk-card layout
• speakers/*.yml — one file per quarter; the single source of truth for the schedule
• ejs/speaker.ejs — template that renders each talk card
• images/ — logo, page mastheads, speaker photos; see images/README.md for the art direction and how to regenerate them
• tools/ — the generators that produce the artwork and bundle the fonts. The mastheads and logo are generated, not hand-drawn: change the parameters and re-run rather than editing SVG by hand
• fonts/ + fonts.css — bundled webfonts, so the site needs no CDN
• js/convergence.js — the home page's live hero animation