You're reading this on a Laravel application. There's no WordPress installation, no headless CMS subscription, no Contentful API key. Every blog post, portfolio item, and service page on this site is a Markdown file in a Git repository \u2014 and the whole thing runs on vanilla Laravel with a custom content pipeline I built from scratch.<\/p>\n
This is the complete technical walkthrough of how it works.<\/p>\n
The first question any Laravel developer asks when building a content site is: what's the database schema?<\/em> I asked it too \u2014 and after sketching out the tables, I stopped and asked a different question: do I actually need one?<\/em><\/p>\n For a content site, the write path is almost never the bottleneck. I publish a few posts a week at most. The read path<\/strong> is everything. And for reads, a flat-file system with aggressive caching is hard to beat:<\/p>\n The trade-off is query flexibility \u2014 but when your content structure maps directly to directories, you don't need The system has three layers:<\/p>\n Every content file is a The frontmatter is parsed by The With this pattern: the first<\/strong> request acquires the lock and populates the cache. Every subsequent concurrent request blocks for up to 2 seconds<\/strong> until the lock is released, then finds a warm cache entry. No stampede, no database \u2014 just one file read per cache cycle.<\/p>\n The routes are straightforward. Each content type maps to a controller:<\/p>\n Controllers stay thin, delegating entirely to This is where the zero-database approach shows its elegance. Because Three feed formats are available at The sitemap dynamically includes all published blog posts, portfolio items, and service pages \u2014 always in sync with the content directory, no separate An Dynamic OG images are generated via Content updates and code changes deploy the same way \u2014 Step 7 is the key: 1. Build a content watching mechanism for local dev.<\/strong> Currently, editing a Markdown file in development requires manually clearing the cache. A simple 2. Add a lightweight CLI for content management.<\/strong> 3. Consider search with a flat index.<\/strong> The current search works by filtering cached collections in PHP \u2014 fine for the current volume. As content grows past a few hundred files, a pre-built Fuse.js JSON index or a lightweight MeiliSearch instance would be more scalable.<\/p>\n A database is the right tool for a lot of problems. A personal portfolio isn't one of them.<\/p>\n The flat-file CMS approach gives content version control for free, eliminates the database migration burden on deploys, makes local development instant, and performs faster under caching than most database-backed CMSes \u2014 at the cost of query flexibility that you won't miss on a content-heavy but write-light site.<\/p>\n The If you're curious about any layer in more depth \u2014 the feed generation, the schema.org implementation, or the Deployer pipeline \u2014 get in touch<\/a> or follow along on the blog.<\/p>\n","date_published":"2026-03-28T12:00:00+00:00","author":{"name":"Michael K. Laweh"},"attachments":[{"url":"https:\/\/klytron.com\/assets\/images\/blog\/laravel-markdown-cms.png","mime_type":"image\/*"}]},{"id":"https:\/\/klytron.com\/blog\/ai-integration-strategy-business-outcomes","url":"https:\/\/klytron.com\/blog\/ai-integration-strategy-business-outcomes","title":"Why Your Business Needs an AI Integration Strategy (Not Just an AI Tool)","summary":"Everyone is adding AI tools. Very few are getting AI results. The difference between a business that uses AI and a business that is transformed by it comes down to one thing: strategy over selection.","content_html":" There is a predictable pattern in how businesses approach AI adoption in 2025.<\/p>\n It starts with a demonstration. Someone sees ChatGPT summarise a document, or Copilot autocomplete a script, and the response is: "we need to get AI."<\/strong><\/p>\n So they get AI. They sign up for a tool. They run workshops. They generate a lot of excitement.<\/p>\n And six months later, two things have happened: the excitement has faded, and nothing material has actually changed in how the business operates.<\/p>\n This isn't a technology failure. It's a strategy failure. And it's the most common and costly mistake I see in AI adoption today.<\/p>\n Buying an AI tool is the equivalent of buying a high-end machine tool for a workshop and leaving it in the corner. The value is theoretical until someone designs a process around it.<\/p>\n AI integration \u2014 real AI integration \u2014 is the work of understanding your business processes, identifying where intelligent automation creates genuine leverage, designing the system that delivers it, and measuring the outcome.<\/p>\n That is strategy. Tools are what strategy selects.<\/p>\n After integrating AI workflows across multiple client engagements, I've identified four categories where the value is clearest and most measurable.<\/p>\n Every organisation has processes where humans are doing work that is structurally repetitive: extracting data from documents, drafting templated communications, categorising inbound requests, generating first-draft reports from structured data.<\/p>\n These are poor uses of human attention \u2014 and near-perfect applications for LLM-powered automation.<\/p>\n Example:<\/strong> A client processing 200+ supplier invoices per month manually, extracting line items, matching against purchase orders, and flagging discrepancies. An LLM pipeline with structured output reduced this from a 3-day monthly task to a 20-minute review of exception cases.<\/p>\n The humans didn't lose jobs. They stopped spending working days on data entry and started spending them on vendor relationships.<\/p>\n In knowledge-intensive businesses, the gap between raw information and an informed decision is often days or weeks. AI shortens this gap dramatically.<\/p>\n This is the domain of Retrieval-Augmented Generation (RAG)<\/strong> \u2014 AI systems grounded in your proprietary documents, structured data, and institutional knowledge. The result is a system that answers questions your team has always had to research manually.<\/p>\n For technology companies and internal IT teams, AI-assisted development is no longer experimental \u2014 it is the new baseline.<\/p>\n Agentic software engineering<\/strong> \u2014 where AI agents plan implementation steps, write code, run tests, and iterate based on test results \u2014 is compressing development timelines by 40\u201360% on well-scoped tasks.<\/p>\n I have integrated agentic development workflows into my own practice to the point where I routinely deliver in days what previously took weeks. The quality bar is higher, not lower \u2014 AI agents are meticulous at edge case coverage in a way that humans operating under time pressure are not.<\/p>\n Beyond internal automation, AI creates competitive differentiation in customer experience. Not chatbots with scripted responses \u2014 but genuinely intelligent interfaces that understand context, remember conversation history, and escalate appropriately.<\/p>\n The businesses winning with AI today are not the ones using it to cut costs. They're using it to deliver a service experience that was previously impossible at their scale.<\/p>\n Before selecting any tool, run this three-step audit:<\/p>\n Step 1 \u2014 Map your high-volume processes.<\/strong> Where do humans spend time doing work that follows predictable patterns? Document the input, the transformation, and the output.<\/p>\n Step 2 \u2014 Identify the decision layer.<\/strong> For each process, ask: where does a human judgement call occur? AI should handle everything below that line. Humans should handle everything above it.<\/p>\n Step 3 \u2014 Define the measurement.<\/strong> What does a 50% reduction in processing time mean in cost terms? What does eliminating error rate mean in rework cost? If you cannot define what success looks like in business terms, you are not ready to implement.<\/p>\n The AI landscape is moving so fast that the specific tool you select today may be superseded in six months. GPT-4, Claude, Gemini, Mistral \u2014 the frontier is shifting constantly.<\/p>\n What does not change is your business process. The organisation that has clearly mapped where AI creates value, defined the data flows, built the integrations, and measured the outcomes \u2014 that organisation benefits from every improvement in the underlying models automatically.<\/p>\n The organisation that bought a tool without a strategy has to restart this work every time the tool changes.<\/p>\n\n
git push<\/code> and Deployer handles everything.<\/li>\ngit log<\/code>.<\/li>\nphp artisan migrate --seed<\/code>.<\/li>\nWHERE category = 'laravel' ORDER BY date DESC<\/code>. You just read the blog\/<\/code> directory and filter a Collection.<\/p>\n
\nThe Architecture: Flat Files + ContentService + Cache<\/h2>\n
resources\/content\/ \u2190 Source of truth (Markdown files)\n \u2193\nContentService \u2190 Parses, validates, transforms\n \u2193 \nLaravel File Cache (1h TTL) \u2190 Serves all requests\n<\/code><\/pre>\n.md<\/code> file with YAML frontmatter:<\/p>\n\n---\n\ntitle: 'Post Title'\nslug: my-post-slug\ncategory: 'Laravel'\ntags:\n - laravel\n - php\nstatus: published\npublished_at: '2026-01-15 09:00:00'\nauthor: 'Michael K. Laweh'\nread_time: 8 min read\n\n---\n\n# Markdown content starts here\n<\/code><\/pre>\nspatie\/yaml-front-matter<\/code> and the body is rendered by league\/commonmark<\/code> with the GitHub-Flavoured Markdown extension, giving me fenced code blocks, tables, task lists, and strikethrough out of the box.<\/p>\n
\nContentService \u2014 The Core Engine<\/h2>\n
ContentService<\/code> is a singleton registered in AppServiceProvider<\/code> that handles all content loading:<\/p>\n<?php\n\ndeclare(strict_types=1);\n\nnamespace App\\Services;\n\nuse Illuminate\\Support\\Collection;\nuse Illuminate\\Support\\Facades\\Cache;\nuse League\\CommonMark\\CommonMarkConverter;\nuse Spatie\\YamlFrontMatter\\YamlFrontMatter;\n\nclass ContentService implements ContentServiceInterface\n{\n public function __construct(\n private readonly string $contentPath,\n private readonly CommonMarkConverter $converter,\n private readonly int $cacheTtl = 3600\n ) {}\n\n public function get(string $path): array\n {\n $cacheKey = 'content.' . str_replace('\/', '.', $path);\n\n return Cache::lock($cacheKey . '.lock', 5)\n ->block(2, function () use ($cacheKey, $path) {\n return Cache::remember($cacheKey, $this->cacheTtl, function () use ($path) {\n return $this->parse($path);\n });\n });\n }\n\n public function all(string $directory): Collection\n {\n $cacheKey = 'content.dir.' . str_replace('\/', '.', $directory);\n\n return Cache::remember($cacheKey, $this->cacheTtl, function () use ($directory) {\n $path = $this->contentPath . '\/' . $directory;\n\n return collect(glob($path . '\/*.md'))\n ->map(fn (string $file) => $this->parse(\n $directory . '\/' . basename($file, '.md')\n ))\n ->filter(fn (array $item) => ($item['status'] ?? '') === 'published')\n ->sortByDesc('published_at')\n ->values();\n });\n }\n\n private function parse(string $path): array\n {\n $fullPath = $this->contentPath . '\/' . $path . '.md';\n\n abort_unless(file_exists($fullPath), 404);\n\n $document = YamlFrontMatter::parseFile($fullPath);\n\n return array_merge($document->matter(), [\n 'content' => $this->converter->convert($document->body())->getContent(),\n ]);\n }\n}\n<\/code><\/pre>\nThe Atomic Lock Pattern<\/h3>\n
Cache::lock()<\/code> call is critical. Without it, a sudden spike of traffic hitting an empty cache simultaneously would trigger a cache stampede<\/strong> \u2014 every request reads the file and writes the cache entry at the same time, multiplying the disk I\/O and converting what should be a cache hit into a thundering herd.<\/p>\nCache::lock($cacheKey . '.lock', 5)\n ->block(2, function () use ($cacheKey, $path) {\n return Cache::remember($cacheKey, $this->cacheTtl, function () use ($path) {\n return $this->parse($path);\n });\n });\n<\/code><\/pre>\n
\nRouting & Controllers: Clean Separation<\/h2>\n
\/\/ routes\/web.php\n\nRoute::get('\/blog', [BlogController::class, 'index'])->name('blog.index');\nRoute::get('\/blog\/{slug}', [BlogController::class, 'show'])->name('blog.show');\nRoute::get('\/blog\/category\/{category}', [BlogController::class, 'category'])->name('blog.category');\n\nRoute::get('\/portfolio', [ProjectController::class, 'index'])->name('portfolio.index');\nRoute::get('\/portfolio\/{slug}', [ProjectController::class, 'show'])->name('portfolio.show');\n<\/code><\/pre>\nContentService<\/code>:<\/p>\npublic function show(string $slug, ContentServiceInterface $contentService): View\n{\n $post = $contentService->get("blog\/{$slug}");\n\n $relatedPosts = $contentService->all('blog')\n ->where('category', $post['category'])\n ->where('slug', '!=', $slug)\n ->take(3);\n\n return view('blog.show', compact('post', 'relatedPosts'));\n}\n<\/code><\/pre>\n
\nFeeds, Sitemap & Discovery<\/h2>\n
ContentService::all()<\/code> returns a sorted Collection<\/code>, generating an RSS feed or sitemap is just a collection transform \u2014 no ORM, no query builder, no N+1 paranoia.<\/p>\nRSS \/ Atom \/ JSON Feeds<\/h3>\n
\/\/ FeedController::rss()\n$posts = $this->contentService->all('blog')->take(20);\n\nreturn response(\n view('feeds.rss', compact('posts'))->render(),\n 200,\n ['Content-Type' => 'application\/rss+xml; charset=UTF-8']\n);\n<\/code><\/pre>\n\/feed\/posts<\/code> (Atom), \/feed\/posts\/rss<\/code> (RSS 2.0), and \/feed\/posts\/json<\/code> (JSON Feed 1.1). Auto-discovery <link><\/code> tags in <head><\/code> allow feed readers and browsers to find them without configuration.<\/p>\nXML Sitemap<\/h3>\n
sitemap_entries<\/code> table needed:<\/p>\n\/\/ SitemapController::index()\n$posts = $this->contentService->all('blog');\n$projects = $this->contentService->all('portfolio');\n$services = $this->contentService->all('services');\n\nreturn response(\n view('sitemap.index', compact('posts', 'projects', 'services'))->render(),\n 200,\n ['Content-Type' => 'application\/xml']\n);\n<\/code><\/pre>\nOpenSearch<\/h3>\n
\/opensearch.xml<\/code> descriptor allows users to add klytron.com as a search engine in their browser, enabling direct site search from the address bar.<\/p>\n
\nSEO: Schema.org & Open Graph<\/h2>\n
SeoService<\/code> generates structured data on every page request, with page-type-specific schemas injected as JSON-LD:<\/p>\n\/\/ For a blog post\n{\n "@context": "https:\/\/schema.org",\n "@type": "Article",\n "headline": "Post Title",\n "author": { "@type": "Person", "name": "Michael K. Laweh" },\n "datePublished": "2026-01-15",\n "image": "https:\/\/klytron.com\/assets\/images\/blog\/hero.png"\n}\n<\/code><\/pre>\nOgImageController<\/code> using PHP's GD library \u2014 branded background, post title, site domain \u2014 all server-rendered as PNG. No third-party image service, no API key, no recurring cost.<\/p>\n
\nDeployment: Zero-Downtime Atomic Releases<\/h2>\n
git push<\/code>:<\/p>\n# My deploy command (via php-deployment-kit)\ndep deploy production\n\n# What Deployer does:\n# 1. Clone latest commit into releases\/YYYYMMDDHHII\/\n# 2. composer install --no-dev --optimize-autoloader\n# 3. npm run build (Vite)\n# 4. php artisan config:cache\n# 5. php artisan route:cache\n# 6. php artisan view:cache\n# 7. ln -sfn releases\/YYYYMMDDHHII current \u2190 atomic swap\n# 8. php artisan cache:clear\n# 9. Clean up old releases (keep last 5)\n<\/code><\/pre>\nln -sfn<\/code> is an atomic operation at the kernel level. The Nginx symlink swap happens in a single syscall \u2014 there's no window where the server is pointing at a broken or partial build.<\/p>\n
\nWhat I'd Do Differently<\/h2>\n
inotifywait<\/code> watcher or a Vite plugin to send a cache-clear signal would smooth this.<\/p>\nphp artisan content:new blog "My Post Title"<\/code> would be cleaner than manually copying frontmatter boilerplate each time.<\/p>\n
\nThe Takeaway<\/h2>\n
ContentService<\/code> described here is the exact code running this site. If you want to build something similar, the architecture is straightforward enough to port to any Laravel project in an afternoon. The key patterns \u2014 atomic locking, collection-based filtering, type-checked frontmatter \u2014 translate directly.<\/p>\nThe Tool vs. Strategy Distinction<\/h2>\n
Where AI Integration Actually Creates Business Value<\/h2>\n
1. Eliminating High-Volume, Low-Judgment Work<\/h3>\n
2. Compressing the Knowledge-to-Decision Gap<\/h3>\n
\n
3. Accelerating Software Development Cycles<\/h3>\n
4. Building Intelligent Customer-Facing Experiences<\/h3>\n
The Framework: How to Audit Your Business for AI Leverage<\/h2>\n
Why Strategy First Matters More Than Tool Selection<\/h2>\n
\n