AI, LLMs & Agents

Chapter 6 of 6 — AI Cloud Engineer Roadmap

This is the capstone. Everything from Chapters 1–5 — the VPC and IAM, the Kubernetes cluster, the Terraform, the CI/CD pipeline, the observability stack — comes together to deploy one real AI project: a text-to-SQL RAG agent, running on infrastructure you built yourself.

What you'll build: a Retrieval-Augmented Generation (RAG) agent that answers natural-language questions by generating and executing SQL, deployed to the EKS cluster from Chapter 2.

Tools: pgvector, LLM APIs, Ollama

Where AI helps: this chapter is the AI project — there's no "AI writes it, you review it" split here. What you still own: the guardrails. SQL safety (no agent should be able to run DROP TABLE because a prompt told it to), retrieval quality, and what happens when the model is confidently wrong.

Modules in this chapter

• How I Use LLMs, by Andrej Karpathy — a working mental model for what LLMs are actually good at
• AI Agent Skills, by IBM
• AI Agents & MCP — how agents call tools, and what the Model Context Protocol standardizes
• Retrieval-Augmented Generation (RAG) — chunking, embeddings, retrieval, why RAG beats a longer context window for most use cases
• Claude Code Introduction, by Anthropic

Why this matters

A RAG agent that runs in a notebook proves you understand the model. A RAG agent that runs on a Kubernetes cluster, behind a CI/CD pipeline, with monitoring that pages someone when retrieval quality drops — that proves you can ship the whole system. That's the actual skill gap AI cloud engineering roles are hiring for right now.

Reference implementation

db-agent is the open-source version of this capstone — a text-to-SQL AI agent deployed on Kubernetes, with SQL safety guardrails and a full CI/CD pipeline, presented at AAAI-25. It's available as a Streamlit app, a Next.js + FastAPI app, and a native Databricks App, so you can see the same agent built three different ways.

---

You've completed the AI Cloud Engineer Roadmap

Six chapters, one working AI project deployed on infrastructure you built yourself, end to end. If you want to go deeper on the data side specifically — S3, Glue, Athena, and the rest of the AWS data platform stack — the AWS Data Engineer Roadmap picks up from here.

Want to do this live instead of solo? beCloudReady runs a weekly cohort that follows this exact curriculum, with Slack support and access to the TorontoAI founder/recruiter network.