Learn AI Engineering by Hand, Not by Wrapper

What it does

This is a curriculum of self-contained Colab notebooks for backend and full-stack engineers moving into AI Engineer or Forward Deployed Engineer roles. It walks through the full applied-LLM lifecycle—prompting, structured output, tool calling, RAG, agents, fine-tuning, security, and operations—using only raw API calls against the free Groq tier. Each notebook handles its own dependencies and reads API keys from Colab secrets, so you can run everything top-to-bottom without a local environment or a credit card.

The interesting bit

The course is intentionally framework-free: you build the agent loop, RAG pipeline, and eval harness from scratch before touching LangChain or LlamaIndex, which means you learn what the wrappers actually do and when to avoid them. Evaluations are the spine of the curriculum, with a “measure before you tune” habit installed before you build anything complex. The final modules are real case studies—debugging a support assistant in production, running a pipeline-vs-agent cost showdown, and surviving a red-team robustness benchmark—rather than toy demos.

Key highlights

• Runs entirely on the free Groq API; no credit card required, with optional Colab-GPU appendices for LoRA fine-tuning and serving concepts
• Covers agents from scratch, MCP, tool design, guardrails, and the SKILL.md progressive-disclosure pattern
• Evals-first curriculum with golden sets, LLM-as-judge, and regression evals treated as CI
• Includes prompt-injection defense, LoRA/QLoRA trade-offs, and LLMOps observability
• OpenAI-compatible patterns throughout, so the skills transfer directly to OpenAI and Anthropic with minimal changes

Caveats

• LoRA fine-tuning and self-hosted serving are concept-first with optional GPU appendices, because Groq does not host training or serving workloads

Verdict

Backend and full-stack engineers interviewing for AI Engineer or Applied AI roles should start here; if you are looking for a no-code quickstart or a LangChain reference manual, this is not it.