AI Engineering
Nine courses for engineers who know Python and want to build reliable LLM applications. Practice APIs, structured output, retrieval, agents, evaluations, safety, infrastructure, and end-to-end design.
The full curriculum is mapped below; courses ship over the coming weeks.
Who this is for
Built for backend, full-stack, ML, and platform engineers who are comfortable with Python. No machine-learning research background is required.
What you'll learn
Reliable LLM integration
Work with APIs, streaming, structured output, retries, fallbacks, and traceable failures.
Retrieval and tool use
Ground responses in source material and design tools that support controlled, multi-step work.
Evaluation and improvement
Create useful baselines and evaluations, then decide when prompting, retrieval, or fine-tuning should improve the system.
Safety and control
Plan for prompt injection, validate outputs, apply guardrails, and keep human review where it matters.
Performance and cost
Reason about latency, caching, observability, versioning, and the tradeoffs behind production infrastructure.
End-to-end system design
Bring models, retrieval, tools, evaluations, and safeguards together in a coherent application design.
More outcomes unlock as later courses ship.
The learning path
Build core capabilities, strengthen quality and safety, then bring the complete system together.
From quickstart to production: streaming, tool calls, structured output, retries, observability.
Chunking, embeddings, vector store, retrieval, reranking, and the eval loop that proves it works.
Tool use, planning, and multi-step orchestration. Build an agent that actually finishes the task, and knows when to stop.
The plan-act-observe loop in depth: stopping conditions, retries, memory, and context budgets that keep an agent on track.
Design rubric-based evals, calibrate judges against humans, and use them to gate releases.
When to fine-tune versus RAG versus prompt, and how to actually do it: data prep, LoRA, evaluation, distillation for cost.
Model registries, deployment patterns, GPU economics, retrieval infra, and the cost lines that decide whether an AI feature ships.
Prompt injection, output validation, PII, and content safety. Ship LLM features that hold up to hostile input.
Pattern recognition for which workflows in your org are worth automating with AI and which aren't.
Put it all together: API, retrieval, an agent, evals, guardrails, and a deploy. One real feature, end to end.