pydata-london-2026

A fun gimmick to promote PyData London 2026 (5–7 June 2026, Convene Sancroft, London).

Tell your LLM to fetch https://pydata.london/recommender.md and follow the instructions and it becomes a personalised conference recommender: top picks, a day-by-day itinerary, and a "stretch" pick to break you out of your bubble.

• 🌐 Landing page: https://pydata.london
• 📄 The artifact LLMs fetch: https://pydata.london/recommender.md
• 🎟 Tickets: https://ti.to/pydata/pydatalondon26

How it works

The whole thing is one static markdown file with two halves:

1. Prompt preamble — instructs the LLM how to behave: ask three multiple-choice questions if it has no memory of the user, then output four sections (top 5 / itinerary / stretch picks / ticket CTA).
2. Schedule data — every session from the official pretalx export, grouped by day, with abstracts and speaker bios.

A daily GitHub Action re-fetches the pretalx export and rebuilds the file, so it stays in sync as the schedule moves.

Layout

index.html                          Landing page (QR, Open-in-ChatGPT button)
recommender.md                      Generated — LLMs fetch this
schedule.json                       Cached pretalx export (committed for diff visibility)
qr.svg                              Generated — QR pointing at the landing page
templates/recommender.md.j2         Jinja template — edit the prompt here
scripts/build_recommender.py        Fetch + render
.github/workflows/refresh-schedule.yml   Daily cron + manual trigger
CNAME                               Custom domain (pydata.london)

Local build

python -m venv .venv
source .venv/bin/activate
pip install requests jinja2 'qrcode[pil]'
python scripts/build_recommender.py

That writes recommender.md, schedule.json and qr.svg. Open index.html in a browser to preview the landing page.

Editing the prompt

The prompt design is in templates/recommender.md.j2 — edit it, run the build script, commit. The schedule rendering happens at the bottom of the same template; everything above the # Schedule data heading is the prompt itself.

Custom domain

The site is served at https://pydata.london via the CNAME file at the repo root (DNS hosted at Hover, A records pointing at the four GitHub Pages IPs).

The org-level CNAME at london.pydata.org is a separate site (the pydatalondon.github.io repo) and is unrelated to this project.

Credits

• Built with the public pretalx JSON export — no scraping required.
• PyData is a NumFOCUS fiscally sponsored project.