Full-Stack Developer & AI Systems Builder · Bogotá, Colombia

I build at the intersection of LLMs and web infrastructure, shipping real products that use AI agents as first-class components, not just features bolted on.

Self-hosted media server stack fully managed by AI through 25 MCP tools. One command (npx create-mediabox) deploys Jellyfin, Sonarr, Radarr, qBittorrent and more, then an AI assistant takes over. Three deployment modes: local, VPS with auto-HTTPS, Cloudflare Tunnel.

Production tech blog and course platform on Next.js App Router. The entire editorial workflow (publishing posts, managing courses, uploading images, scraping metadata) is handled by a Claude agent through a dedicated MCP backend. No manual CMS.

The AI backbone of TechnoEspacio. A custom MCP server that gives Claude autonomous control over content operations: create, edit, publish, fetch OG data, generate and upload cover images. The CMS becomes an API surface for the agent.

I currently work as an AI Code Evaluator at Scale AI / Outlier, which means I'm deep in code every day: auditing TypeScript and Python outputs, writing evaluation rubrics, building RLHF/SFT datasets for agentic workflows. It just doesn't show up in a contribution graph.

That work gives me a close-up view of how language models fail at code, how agent architectures break under real prompts, and what it takes to build reliable AI systems. It informs how I build everything outside of it.

Frontend   ->  React · Next.js (App Router) · TypeScript · Tailwind · shadcn/ui
Backend    ->  Node.js · Express · Python · REST APIs · GraphQL · MCP Protocol
Databases  ->  MongoDB · PostgreSQL · MySQL
Infra      ->  Docker · Caddy · Cloudflare Tunnel · AWS (basic) · Vercel · OAuth2
AI         ->  RLHF/SFT evaluation · Prompt engineering · MCP agent development

• 🌐 technoespacio.com
• 💼 linkedin.com/in/juan-carlos-mu%C3%B1oz
• 📍 Bogotá, Colombia · English C1