How I Built an AI Agent for my Portfolio (Yabasha.dev) using Laravel & Next.js

As a Tech Lead and AI Engineer, I often tell my colleagues: "The cobbler's children have no shoes." We build sophisticated, high-performance systems for others, but our own portfolio sites often gather dust, lacking the polish we demand in our professional work.

I decided to fix this for my own platform, Yabasha.dev. I didn't just want a blog; I wanted a living playground for my expertise in AI Agents and Full-Stack Architecture.

So, I built an AI Agent to be my "Senior Editor" and "SEO Manager," automating the tedious parts of content creation so I could focus on writing. Here is a deep dive into how I built it using Laravel, Next.js, and OpenRouter.

The Architecture: Hybrid Power

I chose a hybrid stack that leverages the strengths of two ecosystems:

• Backend: Laravel 12 (API, Admin Panel via Filament).
• Frontend: Next.js 15 (Static Generation, React Server Components).
• AI Gateway: OpenRouter (Unified API for OpenAI, Anthropic, Google, and Meta models).

This setup gives me the developer experience and stability of Laravel for data management, while delivering the blazing-fast performance of Next.js for the end user.

The Problem: Friction

Writing a technical article is only 50% of the work. The other 50% is:

1. Writing a compelling meta description.
2. Figuring out the primary and secondary SEO keywords.
3. Generating Open Graph (OG) tags so it looks good on Twitter/LinkedIn.
4. Categorizing the content.
5. estimating reading time and difficulty level.

This friction often prevented me from hitting "Publish".

The Solution: The "Editor" Agent

I created a background service in Laravel (

PostAiService) that acts as an autonomous agent. It doesn't just "summarize" text; it analyzes it with a specific persona.

1. The Brain

The core logic resides in a dedicated service that constructs a complex prompt. I instruct the LLM to act as an "Expert SEO Specialist and Content Editor".

Here is a simplified look at the prompt structure I use:

// app/Services/PostAiService.php protected function buildAnalysisPrompt(string $content): string { return <<<EOT ActasanexpertSEOspecialistandcontenteditor. AnalyzethefollowingblogpostcontentandreturnaJSONobjectwith: 1.`summary_short`:Aconcise1-2sentencehook. 2.`level`:'beginner','intermediate',or'advanced'. 3.`intent`:'guide','opinion','case_study',etc. 4.`og_title`:AcatchyOpenGraphtitleoptimization. 5.`primary_keyword`:ThesinglemostimportantSEOkeyword. 6.`secondary_keywords`:Anarrayofsupportingkeywords. Content: {$safeContent} EOT; }

This ensures that every single article I write has structured, machine-readable metadata that Next.js can easily ingest.

2. The Marketplace of Models

One of the coolest features I built is a dynamic model selector. The AI landscape changes weekly. One week GPT-5.2 is king, the next Claude Opus 4.5 takes the crown, or Gemini 3 Pro offers 2M token context.

I didn't want to hardcode a model.

I built an

AiModel resource in Filament that tracks available models from OpenRouter. But I went a step further—I implemented a

Benchmarking System

.

// app/Services/OpenRouterService.php public function benchmarkModel(AiModel $model): ?float { // We fire a standard test prompt to measure real-world speed $startTime = microtime(true); $response = Http::withToken($this->apiKey) ->post($this->baseUrl.'/chat/completions', [ 'model' => $model->request_id, 'messages' => [['role' => 'user', 'content' => 'Count from 1 to 10.']], ]); // Calculate tokens per second (TPS) // ... $model->update(['actual_speed' => $tps]); }

In my admin panel, I can sort models by Cost and Speed (Tokens/Sec) and toggle the active model for my agent instantly. If a new, cheaper model comes out, I add it, benchmark it, and switch my agent to use it.

The Result: Zero-Friction Publishing

Now, my workflow is simple:

1. I write a rough draft in Markdown.
2. I save it.
3. The PostObserver detects the change and queues a job.
4. My AI Agent wakes up, reads my draft, and fills in all the SEO metadata, summaries, and categorization.
5. Next.js rebuilds the page, pulling this rich metadata into the <head> of the document.

The result is a perfectly optimized page on Yabasha.dev without me spending a single second on "optimization."

Conclusion

AI isn't about replacing engineers; it's about amplifying them. By building this agent, I've removed the administrative burden of blogging, allowing me to focus entirely on sharing knowledge.

If you're interested in building similar AI-driven architectures or need a senior pair of hands on your Next.js/Laravel stack, feel free to reach out or check out my work at Yabasha.dev.