Academic Website Auto-Updater v2

This project now has a lightweight automation layer for keeping the academic website data-driven and reviewable.

What It Does

Imports new publications from arXiv metadata or exported BibTeX.
Reuses the existing publication generator to create publication cards and update topics/links data.
Suggests selected papers from pinned slugs, topics, year, and highlight metadata.
Audits the automation config in CI without making network requests.
Keeps group members, research projects, and news entries data-driven through YAML/front matter templates.
Uses the existing Jekyll feed for RSS and the existing GitHub Pages workflow for deployment.

Safe Source Policy

Google Scholar should not be scraped. If Scholar is the source of truth, export BibTeX manually and import that file as google_scholar_manual.

Preferred sources:

arXiv IDs through the arXiv API
BibTeX exported from DBLP
BibTeX exported from Semantic Scholar
BibTeX exported from OpenReview, publisher pages, or Google Scholar
Hand-reviewed YAML for papers that need custom abstracts, topics, or links

Common Commands

make auto-plan
make auto-audit
make suggest-selected
make sync-publications

All write-capable commands dry-run by default. To write changes:

make sync-publications APPLY=1
make suggest-selected APPLY=1

Import one arXiv paper directly:

ruby scripts/auto_updater.rb import-arxiv 2606.00395 --topics "LLM Systems,MoE Systems,Reinforcement Learning" --apply

Import one BibTeX file:

ruby scripts/auto_updater.rb import-bibtex path/to/paper.bib --topics "LLM Systems,Data & Evaluation" --apply

Configured Sync Queue

Add pending publication sources to _data/auto_updater.yml:

publication_sync:
  sources:
    - kind: arxiv
      id: "2606.00395"
      collection: "technical_reports"
      topics:
        - LLM Systems
        - MoE Systems
        - Reinforcement Learning

Then run:

make sync-publications

Review the dry-run output. If it looks right:

make sync-publications APPLY=1
make validate
make quality

Data-Driven Content

Templates live in scripts/templates/:

publication.yml
publication.bib
news.yml
group-member.yml
research-project.yml

Group members live in _data/group.yml, research project entry points live in _data/research_themes.yml, and news entries live in _news/.

CI/CD

Pull requests run:

auto-updater config audit
publication metadata validation
publication link inventory
Jekyll build
generated-site quality audit

The scheduled link-check workflow still checks publication links. RSS is provided by jekyll-feed, and deployment remains GitHub Pages-first. vercel.json is included as an optional static deployment config for Vercel.

Hongyi Wang