AI Engineering with Claude

Set up your workspace for the course: configure your Anthropic API keys, get Claude Code running, and preview the harness engineering skills you'll build.

Explore the Claude 4.5 model family—Haiku, Sonnet, and Opus—each optimized for a different balance of intelligence, speed, and cost for efficient AI applications.

Learn to select the best Claude 4 model—Haiku, Sonnet, or Opus—using the API, matching model strengths to task complexity for faster, cost-effective, and high-quality AI solutions.

Discover agentic AI: systems that perceive, reason, and act to achieve goals through a perception-reasoning-action loop, core components, and agent design patterns.

Learn to design agent architectures for automating complex tasks using multi-agent systems, enabling speed, specialization, resilience, and scalable AI-powered workflows.

Learn to build autonomous AI agents with the Claude Agent SDK—set goals, assign tools, manage permissions, and automate complex workflows using natural language instructions.

Learn to build production AI agents with Claude Agent SDK that use tools to read, process, and write files, automating complex document workflows with clear prompts and structured outputs.

In this lesson, you'll build a working claims intake agent across three cumulative exercises, starting from a stop_reason-driven agentic loop and ending with a model that decides on its own when to clarify, route, or escalate.

In this lesson, you'll build a context-engineering harness for a long-running retail support copilot — pruning verbose tool output, keeping a persistent block of case facts, compressing resolved conversations on a token budget, and assembling it all so the facts that matter sit where the model reads them best.

Discover how to use Claude Code as an agentic AI assistant—reading your project, writing code, scaling tasks with subagents, and applying custom knowledge via CLAUDE.md and skills.

Learn to build sophisticated AI systems by configuring Claude Code with modular agents and reusable skills for task delegation and domain-specific workflows.

Learn to create and structure Claude skills—modular files that encode expertise—enabling AI agents to perform tasks consistently to your team's standards.

Learn to create and use Claude Skills to give agents reusable expertise. Build agents that discover, combine, and apply skills for complex tasks, with structured, type-safe outputs.

In this lesson, you'll stand up a version-controlled Claude Code configuration for a 12-person retail team working in one monorepo — a modular CLAUDE.md hierarchy with @import standards, path-scoped rules that load per file surface, a read-only /review slash command, a forked /deploy-check skill that keeps its verbose output out of the main session, and a decision doc that names when to use plan mode, direct execution, and the Explore sub-agent. You assemble it across four cumulative exercises, and a Python validator certifies the finished configuration against 35 acceptance tests.

Across four cumulative exercises, you'll assemble a Layer 3 Claude orchestration that wakes up on a schedule, looks at only what changed since last time, and survives a mid-run crash without losing its work: tiered hot/warm/cold state with atomic writes, a push-work-down invocation pipeline, an fsync'd manifest with a resume-vs-fresh recovery rule, and a fork-with-merge scratchpad workflow.

Get oriented before you build: the course prerequisites and outcomes, your instructors, and how to set up your Vocareum workspace and Claude API key.

Understand MCP's role in connecting AI apps to external tools, MCP core components, real-world use cases, and complete a hands-on job search assistant walkthrough.

Explore the MCP architecture, roles of host, client, and server, communication via JSON-RPC, transport mechanisms, and lifecycle phases of an MCP workflow.

Explore the three core MCP server features—tools, resources, and prompts—that enable LLMs to act, access data, and use guided templates for effective user interactions.

Explore key MCP client features: roots (directory restrictions), sampling (LLM requests), and elicitation (user info requests) for secure, flexible workflows.

Learn to build an MCP server that provides calculator tools, mathematical constants as resources, and calculation prompts for AI applications, using FastMCP and best development practices.

Learn to integrate Claude Agent SDK agents with external tools via MCP servers, enabling automated tasks like code analysis and structured data retrieval from services such as GitHub and ESLint.

Assemble a production-minded retail inventory agent by designing structured tool results, differentiated tool descriptions, a tool-distribution and tool_choice policy, and a built-in-vs-custom tool-selection framework.

Learn how to design, build, and integrate custom tools in Claude Agent SDK, letting agents interact with private data, APIs, and automate specialized business tasks.

Learn to create, integrate, and test custom tools in the Claude Agent SDK, enabling agents to interact with APIs and solve real-world tasks tailored to your needs.

Build a Python governance layer that turns the four enterprise MCP decisions into working code, a scoped config registry with a secret-leak CI gate, custom servers with built-in-resistant tool descriptions and a schema resource, and an audited agent loop with an append-only audit trail.

Get oriented before you start: see what the course covers, who it's for, and the outcomes you'll reach, plus the prerequisites and resources you'll use along the way.

Learn how to use schemas to enforce structured outputs in Claude Agent SDK, transforming unpredictable text into reliable, validated data for robust applications.

Learn to implement structured outputs in Claude Agent SDK by defining schemas with Zod, enabling reliable, validated data extraction for robust applications.

In this lesson, you'll build a production-shaped document extraction pipeline with the Anthropic API, one technique at a time. You'll design a JSON Schema that refuses to fabricate, orchestrate a two-pass classify-then-extract flow with forced and `any` tool choice, write an extractor prompt that stops the model from inventing values, and add a validator that catches arithmetic the model would otherwise silently "correct." Along the way you'll build the intuition for when each pattern is the right tool — and when it isn't.

Agent Evaluation guides assessing an agent’s task completion, quality, tool use, and system metrics using response, step, or trajectory strategies to ensure reliable and efficient operations.

Learn to build automated evaluation frameworks for Claude Agent SDK agents, using agent traces, evaluators, test cases, and schema validation to ensure reliable performance.

In this lesson, you'll build a production-shaped document extraction pipeline with the Anthropic API, one piece at a time. You'll learn how to retry the mistakes a model can fix and escalate the ones it can't, batch a monthly workload for a cost discount, add an independent reviewer for a second opinion, and route each policy to the right human queue.

Build a Python system that fuses four disagreeing evidence sources into one honest, provenance-preserving briefing that annotates conflicts, escalates on explicit criteria, and keeps running when a source fails.

Start here for a tour of the course: the prerequisites you'll need, the outcomes you'll reach, and the resources that support building safe, guardrailed agents.

Discover how to design multi-agent AI systems using orchestrators and subagents, apply sequential, parallel, and advanced orchestration patterns, and automate complex, multi-step tasks.

Learn to design and implement multi-agent systems using Claude Agent SDK where an orchestrator coordinates specialized subagents to automate and scale complex, multi-step business processes.

Assemble a multi-agent system that triages electronics-defect reports across four cumulative exercises: scoped subagent definitions, parallel spawn with scoped context, Pydantic-validated structured handoff, and a bounded iterative refinement loop. By the end you have a working CLI that runs against the real Anthropic API.

Wrap an Anthropic-SDK banking agent in a compliance enforcement layer across three cumulative exercises: a programmatic KYC prerequisite gate, a PostToolUse hook that normalizes heterogeneous tool output, and an interception hook that redirects over-threshold transfers with a structured handoff. A comparison harness proves deterministic hooks block 100% of violations where a strong prompt cannot.