Building LLM agents that survive production, and the full stack around them.

I build AI systems that go past the demo:

• agents with a closed harness — policies, hooks, guardrails, evals
• RAG and retrieval that keeps the cost per query survivable
• the application around the agent, from the flow to the deploy

My work sits where agents, product and production reliability meet.

→ agent harnesses
→ MCP and tool use
→ cost and latency per query
→ AI-assisted development that follows the team's architecture

• Conversational AI Engineer and Full Stack Developer at Smartspace
• Moved a large hospital's exam scheduling from n8n to an agent in code — completed bookings went from ~120 to ~230 a month
• Cut an agent's cost per query by 99.3%, from US$ 6.00 to US$ 0.04, with retrieval over Pinecone in place of a 600k-token prompt — n8n-rag-catalog-optimizer
• Built a design-to-bot pipeline that cut a bot's engineering cycle by 80%, from Figma prototype to production bot
• Systems Analysis and Development (CST) at Unipê - Centro Universitário de João Pessoa

Python · FastAPI · TypeScript · Node.js · React · Next.js
LangChain · LangGraph · RAG · Pinecone · MCP · evals · Botpress · n8n
PostgreSQL · MongoDB · Supabase · AWS · Docker · Terraform

An agent that demos well and an agent that operates are separated by one thing: the harness. I would rather ship the second one.

• LinkedIn: https://www.linkedin.com/in/dv-dev/
• Email: danielvmacedog1@gmail.com