How to Build a Modern Website in 2026: Full Tech Stack and Architecture Case Study

When engineering a modern tech website in 2026, it is easy to get trapped in unnecessary complexity: heavy SPA frameworks, dozens of SaaS dependencies, recurring database bills, and bloated client-side bundles slowing down mobile users.

Under the Jak Zbudować Stronę brand, I established a clear architectural standard for paweldomanski.eu: build a platform that is blisteringly fast (100/100 Lighthouse), free from third-party vendor lock-in, strictly GDPR-compliant, and fully optimized for both traditional search and generative AI engines.

Here is the complete deconstruction of the technology stack I built and recommend.

1. Foundation: Astro 7 and Islands Architecture

Traditional React or Next.js applications ship dozens of megabytes of client JavaScript just to display static text. In Astro 7, the core rule is Zero JS by Default.

• Static Site Generation (SSG): All marketing and editorial pages are pre-rendered into pure HTML during build (output: 'static'). Edge time-to-first-byte (TTFB) is consistently below 30ms.
• Astro Islands: Client JavaScript is loaded strictly for components that genuinely require interactivity (such as the contact form or language switcher), and only when they enter the viewport (client:visible).
• MDX: Articles are authored in MDX, combining the speed of Markdown with the power of embedding interactive custom UI components.

2. Content Management: Git-Backed Headless CMS (TinaCMS 3.12)

Instead of a monolithic database-backed CMS like WordPress, the platform uses TinaCMS:

• Zero Database Lock-in: All content (14 collections: articles, brands, tech stack, services, case studies) is version-controlled as .mdx and .json files directly inside Git.
• Real-Time Visual Editing: Editors access an intuitive WYSIWYG editor with live preview at /admin.
• Git Versioning: Every content edit produces a Git commit, offering free backup, instant rollback capability, and complete audit history.
• Deployment Freedom: Runs locally without authentication, via TinaCloud, or on a self-hosted Docker container.

3. Zero-Subscription Search: Pagefind (WASM)

Instead of paying $50–$200/month for Algolia or Typesense, I integrated Pagefind — a static WebAssembly search engine:

• Indexes articles, services, and technologies in milliseconds during the build process (scripts/build-search.js).
• Generates micro-chunks of index data that the browser downloads only when the visitor begins typing.
• Provides native multi-language support (Polish and English) with faceted filtering across topic pillars.

4. Accessibility (WCAG 2.2 AA) and Core Web Vitals (0.000 CLS)

Accessibility and visual stability are non-negotiable engineering standards:

• Strict Heading Hierarchy: Exactly one <h1> per page, no skipped heading levels (logical H2 -> H3 structure) verified by automated auditing.
• Keyboard-First Navigation: Full Tab navigation, high-contrast focus-visible outlines, semantic form labels, and aria-invalid state indicators.
• Layout Stability (CLS = 0.000): Fixed-position overlays and zero DOM shifting ensure top-tier Google PageSpeed scores.

5. Privacy & Regulatory Compliance: Google Consent Mode v2

Privacy compliance under European law (GDPR/ePrivacy) is built into the architecture:

• Compact Cookie Banner: Occupies under 15% of mobile viewport height to prevent Google intrusive interstitial penalties, running non-blocking vanilla JS on requestIdleCallback.
• Google Consent Mode v2: GA4 tags (G-F4D5VGQ5L1) default to analytics_storage: 'denied' until explicit affirmative user consent is granted.
• Privacy-First Self-Hosted Analytics: In parallel, DexterLab Analytics (stat.dexterlab.pl) provides anonymous visitor tracking with no tracking cookies.
• Bilingual Legal Suite: Tailored privacy policy, cookie policy, terms of service, and returns framework (PL & EN).

6. Generative Engine Optimization (GEO): AI Search Readiness

In 2026, discovery encompasses ChatGPT, Claude, Perplexity, and Google AI Overviews. To ensure autonomous AI crawlers index content accurately:

• llms.txt and llms-full.txt: The official open standard (llmstxt.org) providing structured, markdown-first documentation for LLMs.
• JSON-LD Schema Markup: Comprehensive Schema.org graphs on every page (WebSite, Person, TechArticle, Service, BreadcrumbList) establishing named authorship.
• AI-Optimized robots.txt: Explicit crawl permissions for GPTBot, ClaudeBot, PerplexityBot, and Applebot.
• Bilingual RSS Feeds: Machine-readable syndication at /rss.xml and /en/rss.xml.

7. Multi-Cloud Infrastructure: Cloudflare Workers & Vercel

The build pipeline is provider-agnostic:

• Cloudflare Workers (Static Assets): Configured via wrangler.jsonc to deploy static assets directly to Cloudflare's global edge network with zero server overhead.
• Vercel Build Output API v3: Integration via @astrojs/vercel allows parallel zero-config deployment on Vercel without 404 errors.
• Node Container Fallback: Capable of running inside Docker on bare-metal servers or VPS via @astrojs/node.

Key Takeaways for Technical Founders and CTOs

Modern web engineering is not about adding more dependencies — it is about deliberate simplification. Combining Astro 7, Git as a database (TinaCMS), and client-side WebAssembly (Pagefind) creates a system that loads instantaneously, costs pennies to run, and is immune to external database outages.

If you are planning to modernize your web architecture or need a performance audit, explore my advisory services or schedule an architectural consultation.