I Built an AI SEO Brief Generator That Compares 3 Models Side-by-Side — Here's the Architecture

I spent 3 months building an SEO content brief generator, and I wanted to share the technical architecture with the dev.to community.

The Problem

SEO content briefs are the pre-writing research phase for blog posts — you analyze the SERP, figure out search intent, outline the article structure, and suggest metadata. Most people do this manually, spending 1-2 hours per article.

The Solution

seobrief.cc — keyword in, complete brief out in ~30 seconds.

Technical Architecture

Frontend: Next.js 16 + Tailwind v4 + React Server Components

Backend: Next.js API routes (edge-compatible)

Database/Auth: Supabase (Postgres + Auth + Row Level Security)

Payments: Stripe Checkout (subscription)

AI: 3 LLMs running in parallel:

• DeepSeek V3 (primary, best quality)


• Qwen (backup, fast)


• Moonshot (backup, different perspective)

Hosting: Vercel (production), with @vercel/analytics for traffic

How the Multi-Model Comparison Works

typescript

// Simplified — the actual route calls 3 providers in parallel

const [deepseek, qwen, moonshot] = await Promise.allSettled([

  generateBrief(keyword, 'deepseek'),

  generateBrief(keyword, 'qwen'),

  generateBrief(keyword, 'moonshot'),

]);



// Score each result, return all 3 + recommendation

const results = [deepseek, qwen, moonshot]

  .filter(r => r.status === 'fulfilled')

  .map(r => ({ ...r.value, score: scoreBrief(r.value) }));



return { results, recommended: results.sort((a, b) => b.score - a.score)[0] };