Technology
- Framework
- Next.js 16 (App Router, static export)
- Language
- TypeScript in strict mode
- UI
- React 19 (server components by default)
- Styling
- Tailwind CSS 4 with a small set of design tokens
- Content
- MDX with gray-matter frontmatter; rehype-slug + autolink + KaTeX + selective highlight.js
- Search
- Build-time JSON index served at
/search-index.json; pure-JS scorer client-side - Feeds
- RSS 2.0 at
/rss.xml, JSON Feed 1.1 at/feed.json - Hosting
- nginx serving a static export; deployed via SSH-CA-signed automation
- Build artifacts
- Per-article OG images via
next/og; auto-generated tag landing pages; sitemap.xml - Offline reading
- Service worker at
/sw.jscaches every HTML page you visit so previously-read articles work without a connection - Bulk citation
- The full library is exposed as BibTeX at
/citations.bib; per-article APA/Chicago/BibTeX/plain in the article sidebar
Typography
- Serif (headings)
- Lora — designed by Olga Karpushina & Cyreal
- Sans (body)
- Inter — designed by Rasmus Andersson
- Monospace (code)
- JetBrains Mono — JetBrains Type Lab
- Delivery
- woff2-only, self-hosted via @fontsource; 9 faces total
Standards
- Accessibility
- Target: WCAG 2.2 AA (contrast, focus visibility, motion preferences, keyboard navigation, ARIA semantics)
- Security headers
- HSTS, CSP without
unsafe-eval, X-Frame-Options DENY, COOPsame-origin, CORPsame-origin, Permissions-Policy denying 25 features - Performance
- Static-only HTML/CSS/JS; immutable caching on hashed assets; preloaded critical fonts; tree-shaken icon set
- Search engines
- Per-page canonical URLs, Article + BreadcrumbList JSON-LD on every article, sitemap.xml indexed weekly
- AI crawlers
- Explicit allow-list in robots.txt for GPTBot, ClaudeBot, PerplexityBot, etc. — research belongs in retrieval indexes
Editorial
- Sourcing
- Every article is researched against primary academic literature; claims link to journal articles, conference proceedings (ICDAR, CVPR, NeurIPS), or authoritative documentation where available.
- Fact-checking
- Quantitative claims, citations, historical references, and technical specifications pass through a documented fact-checking pipeline before publication.
- Tone
- Plain academic prose. Hedging where the evidence is mixed. No marketing language; no fabricated benchmarks or hype.
- Updates
- OCR research moves quickly. Articles carry both a
publishedAtand an optionalupdatedAtfield; the latter surfaces as an “Updated” badge in the article header when present.
Found a problem?
Technical issues, factual corrections, accessibility regressions — all are welcome. We’d rather know than not. Reach us through the channels on the about page.