Content and build pipeline for https://s-anand.net/

Content:

• pages/: Standalone pages as Markdown (pages/slug.md).
  • Home page
  • Pages can be nested: pages/lists/slug.md
• posts/: Blog posts as Markdown (posts/yyyy/slug.md).
• assets/: Converted media files used by posts (WebP/OPUS). Served at /blog/assets/.

Configuration & Code:

• metadata.yml for taxonomies (categories, tags), and author info
• hugo.toml: Hugo site configuration
• setup.sh: Build script to generate content and build site
• .github/workflows/deploy.yml: Deployment workflow for GitHub Pages.
• layouts/: Hugo layout overrides for theme customizations.
  • layouts/partials/head.html generates <title> by capitalizing (e.g. blog → Blog) and deduplicating site title (e.g. "S Anand | S Anand" -> "S Anand").
• static/: Static files (CSS overrides, favicon assets).
• themes/PaperMod/: Hugo theme sources (vendored).
• scripts/: Conversion and utility scripts.
• justfile: Justfile for local pre-processing (e.g. analysis).
• analysis/: Data analysis scripts and results (e.g. embedding analysis).

Auto-generated (DO NOT edit!):

• content/: Contains Hugo content (posts/pages + taxonomy and archive index pages).
• public/: Build output (deployed to GitHub Pages).

After editing source files, rebuild with:

bash setup.sh

This runs:

1. scripts/build_content.py - generates content/ from posts/, pages/, and metadata.yml
2. hugo - builds static site to public/blog/
3. Post-processing scripts for comments and feed normalization
4. Copies special pages to public/ root

GitHub automatically runs setup.sh on push to main and deploys public/ to GitHub Pages.

WIP commits are pushed to the live branch. The prod branch holds permanent changes - in case of rollbacks to main.

content/embeddings.parquet contains Gemini Embedding 2 vectors for every page and post.

scripts/embeddings.py                    # embed all new/changed files
scripts/embeddings.py --since 2025-01   # only files modified after a date
scripts/embeddings.py --limit 10        # test run: at most 10 files
scripts/embeddings.py --force           # re-embed all, ignoring hashes

• Model: gemini-embedding-2-preview, 768 dimensions, RETRIEVAL_DOCUMENT task
• Each file's title (from frontmatter) is prepended to the body before embedding
• State is persisted in content/embeddings.duckdb so interrupted runs resume automatically — only files whose content hash changed are re-embedded

• classes: wrap-code adds the wrap-code class to the post's main <article> element, which applies CSS to wrap long code blocks.
• build: { list: never, render: always } ensures that posts/pages are not listed anywhere blog index but are still rendered.
• robotsNoIndex: true adds a <meta name="robots" content="noindex"> tag to the page header to prevent indexing by search engines.
• aliases: ["old-path"] adds redirects from old-path to the current page using Hugo Aliases.