Skip to content

Public project

Content-Driven Church Website

A bilingual site for a local church in Bitola: 2,500+ prerendered routes over a CMS and a build-validated Bible kept in git.

Visit the live site

  • Next.js
  • React Server Components
  • Sanity CMS
  • Zod
  • Fuse.js
  • Vercel

Context

A bilingual website, in Macedonian and English, for a local church in Bitola: editorial pages, sermons, books, a gallery, and a full Bible reader with search, 66 books, 1,189 chapters and 31,102 verses per locale.

My role

Sole engineer: design, build, launch and operation. Planned in May, built June to July 2026, live since July 2026. Operated with Vercel Analytics, Speed Insights and a monthly automated health check.

Challenge

Two kinds of content with opposite needs share one site. Editorial content must go live when the editor publishes, without a developer or a rebuild. The Bible is large, rarely changes and must never be wrong. The answer: a statically generated front end over two independent sources, with no database and the CMS never in the request path.

Constraints

  • A non-technical editor maintains the site alone.
  • Near-zero running cost.
  • It must keep working untouched for years.
  • Bilingual interface and Bible; sermons and books in Macedonian.

Architecture

No database. The CMS is never in the request path.

Screens

  • The Macedonian home page on desktop: a welcome heading, buttons to read the Bible and get in touch, and a hero image of an open Bible with links to the Bible, sermons, books and community.
  • A Bible chapter on desktop: John 3 in Macedonian, with breadcrumbs, a grid of chapter links and numbered verses.
  • Bible search on a phone: the query "љубов" (love), with matching verses listed by reference and the match highlighted.

Key decisions

  1. Bible text in git, behind a build gate

    The Bible lives in git as JSON, never in the CMS: builds are deterministic, every verse has git history, and no Bible page depends on a running service. Verse text never passes through an LLM. Every build validates it with Zod schemas and cross-file integrity checks, so bad content cannot deploy, and each error names the file and the fix.

    Trade-off
    Correcting a verse needs a git push, not the CMS.
  2. A build-time search index, loaded on focus

    The build generates a flat index of every verse and verifies it maps one-to-one onto them. The browser downloads it, about 1.5 MB compressed, only when the search box gets focus; Fuse.js then searches it with no backend.

    Trade-off
    The first Macedonian search parses about 9.7 MB, building the Fuse index on the main thread.
    Alternative
    Hosted search. Cost, a key, a quota, and an index that can drift from the pages.
  3. On-demand revalidation by webhook

    A Sanity publish calls a webhook that runs revalidateTag for that content type. Pages using that tag are marked stale and regenerate on their next visit, with no rebuild.

    Trade-off
    Nothing expires on a timer: if the webhook breaks, published changes don't reach the site until the next deploy. It's the main operational risk.
    Alternative
    Time-based revalidation. Scheduled regeneration runs even when nothing changed, and content stays stale up to one interval.

Outcome

  • Accessibility, Best Practices and SEO: 100 on the home page, Bible index and a chapter.
  • Performance 92–94 on those pages, with CLS 0.
  • No page-specific JavaScript on the Bible reading path: chapters hydrate nothing of their own.

Measured with Lighthouse (mobile, simulated throttling) on the home page, the Bible index and a Bible chapter.