LangGraph and multi-agent AI labs
Ten labs that take you from a single ReAct loop to a guarded, evaluated multi-agent system.
An AI agent is a language model running in a loop: it decides which tool to call, reads the result, and decides again until the task is done. Everything hard about agents lives in that loop: how it plans, what it remembers, how many agents share the work, how they call tools safely, and how you know it worked. Multiple-choice questions can describe those trade-offs; you only really learn them by building the loop three ways and watching them fail differently.
The collection starts with a free lab (a ReAct research librarian on NVIDIA NIM), then builds the same support agent as ReAct, tool-calling and plan-and-execute and compares them. From there it adds long-term memory with a LangGraph checkpointer plus Milvus, reliable structured output, a LangGraph supervisor routing to specialists, two agents talking over the A2A protocol, and an MCP tool server that a LangChain agent connects to. It closes with the operational side: routing across model tiers to cut cost, an LLM-as-judge evaluation harness, and NeMo Guardrails for jailbreak and topical rails.
All ten labs are hosted (no GPU pod needed), run in the browser, and take 25 to 40 minutes each.
The labs, in order
All labs- 1HOSTEDFree# ReAct · thought/action/observe1. Thought:2. search_docs(…)3. Observation:Final answer → 200 OKncp-aai · react-agent-nimIntermediate
Build a ReAct Agent with NVIDIA NIM
Build a working AI research librarian — an agent that can search a corpus of ML papers, read abstracts, compare methods, and reason over them to answer multi-step questions. Uses LangChain, LangGraph, and NVIDIA NeMo Agent Toolkit on real NIM endpoints.
Freencp-aai35 minHostedLaunch - 2HOSTED Proncp-aai · agent-patternsIntermediate
Build an AI Agent 3 Ways: ReAct vs Tool Calling vs Plan-and-Execute
Build the same SaaS customer support agent three different ways — ReAct, direct tool calling, and plan-and-execute — then compare them on speed, reasoning quality, and reliability to learn when to use each pattern in production.
ncp-aai35 minHostedLaunch - 3HOSTED Pro# Milvus + LangGraph$ checkpoint.saveshort_term ...... 12 msgslong_term ....... 84 factsrecall@5 = 0.92ncp-aai · agent-memoryIntermediate
Add Long-Term Memory to an AI Agent: LangGraph + Milvus
Build a sales intelligence assistant that remembers — short-term conversation state with LangGraph checkpointer, long-term facts in Milvus, and reflection loops that auto-extract knowledge. Learn the memory architecture every production agent needs.
ncp-aai35 minHostedLaunch - 4HOSTED Pro# structured-output-tools · agentPOST /api/agent/invoke200 OK · gradedncp-aai · structured-output-toolsIntermediate
Structured Output & Function Calling with NIM
Get reliable machine-parseable data out of an LLM. Compare prompt-only JSON extraction against the function-calling API, chain two tools, and measure the reliability gap on a real extraction task.
ncp-aai30 minHostedLaunch - 5HOSTED Proncp-aai · multi-agent-orchestrationIntermediate
Build a Multi-Agent Supervisor with LangGraph
Build a supervisor agent that routes queries to specialist agents — a core architecture pattern tested on the NCP-AAI exam.
ncp-aai40 minHostedLaunch - 6HOSTED Proncp-aai · a2a-communicationAdvanced
Build Two Agents That Talk via the A2A Protocol
Build two independent agents that talk to each other via the A2A protocol — each owned by a different team, running in its own process, discovered through a standardized AgentCard. Learn how A2A differs from multi-agent orchestration and when each architecture fits.
ncp-aai40 minHostedLaunch - 7HOSTED Proncp-aai · mcp-tool-serversAdvanced
Build an MCP Tool Server & Connect a LangChain Agent
Build a Model Context Protocol server that exposes your company's tools and data — then connect a LangChain agent to it. Learn how MCP decouples tools from agents, when to use MCP vs Anthropic Skills vs native @tool, and why MCP is the emerging standard for AI tool interop.
ncp-aai40 minHostedLaunch - 8HOSTED Pro# model-routing-cascade · agentPOST /api/agent/invoke200 OK · gradedncp-aai · model-routing-cascadeIntermediate
Model Routing & Cost Cascade with NIM
Save 60–80% on inference by cascading queries through cheap → mid → expensive NIM models. Measure real costs via NIM's usage.cost field and compare against an always-large baseline.
ncp-aai25 minHostedLaunch - 9HOSTED Pro# agent-evaluation · agentPOST /api/agent/invoke200 OK · gradedncp-aai · agent-evaluationIntermediate
Evaluate an Agent with LLM-as-Judge
Build an eval harness that scores agent responses automatically — correctness via a reference-based judge, plus an accuracy metric and A/B comparison. Same pattern used by NeMo Evaluator for production agent evaluation.
ncp-aai30 minHostedLaunch - 10HOSTED Pro# NeMo Guardrails$ rails.check(query)jailbreak ....... BLOCKEDoff-topic ....... BLOCKEDIT query → ALLOWEDncp-aai · safety-guardrailsIntermediate
Build NeMo Guardrails for an AI Agent: Jailbreak & Topical Rails
Build a guarded IT support agent that blocks jailbreaks, refuses off-topic questions, and safely handles IT queries — using keyword checks, LLM-based validation, and NeMo Guardrails.
ncp-aaincp-genl35 minHostedLaunch
Before you start
- Python basics; you will read and edit short scripts, not write a framework
- What a tool call is (the model emitting a function name plus arguments)
- No prior LangGraph experience; the labs introduce nodes, edges and state as they go
Every lab in this collection, plus every practice test on the platform
$29.99 per month or $290 per year. Cancel any time. Free labs need only an account.
Certifications these labs prepare you for
Frequently asked questions
Yes. "Build a ReAct Agent with NVIDIA NIM" is free with an account. The remaining nine labs are included in Preporato Pro along with every practice test on the platform.
They track the NVIDIA NCP-AAI blueprint most closely (agent architecture, cognition and memory, evaluation, safety, human oversight) and the orchestration, integration and governance domains of Anthropic's CCA-F and CCAR-P. Several labs are referenced from those exam guides on this site.
No. The first three labs introduce the ideas (state, nodes, checkpointers) inside the exercise, and every step is checkable, so you find out immediately when your graph is wired wrong.
Each step has an automatic check against the live environment: the agent must actually route the query, the memory must actually persist across turns, the guardrail must actually block the jailbreak. There is no self-assessment.