A 40-hour, weekend-only, live-online program for developers who want to move past single-shot prompting into stateful, tool-using, multi-agent systems — built with LangGraph, LangChain, and Microsoft's own Agent Framework, and shipped on Azure.
Most GenAI courses stop at "call an LLM API." This program starts where agentic engineering actually begins: designing stateful graphs of reasoning and tool-use, grounding them in your own data, keeping them safe, comparing frameworks, and running them reliably in production on Azure — all built, session by session, with your own hands on the keyboard.
New to Python, or rusty? Work through the companion Python Refresher module — basics through agentic-ready patterns, self-paced, ~10h — before Weekend 1.
Structured prompting, function-calling schemas, few-shot & chain-of-thought patterns that hold up under real inputs.
Context windows, chunking, and assembling system / tool / memory / retrieved context under a token budget.
Short- and long-term memory with LangGraph checkpointers, persisted on Azure Cosmos DB / Redis.
Hybrid vector search on Azure AI Search, re-ranking, and grounded, cited answers.
Supervised fine-tuning and model evaluation inside Microsoft Foundry, and knowing when to skip it.
Content Safety, prompt-injection defenses, tracing, evals, and cost/latency monitoring in production.
Every session is 4 hours, split across a focused teaching block and a hands-on lab, so each weekend leaves you with a working piece of an agent stack — not just slides. All sessions are delivered live, online.
| Weekend | Day | Hrs | Theme |
|---|---|---|---|
| W1 | Saturday | 4h | Foundations — Agentic Architecture & Microsoft Foundry LLM apps vs. agentic systems, ReAct & Plan-Execute patterns, the Foundry model catalog |
| W1 | Sunday | 4h | Prompt & Context Engineering Structured output, function calling, context budgeting, prompt evaluation |
| W2 | Saturday | 4h | Memory Management with LangGraph Checkpointers, short/long-term memory, Azure-backed persistence |
| W2 | Sunday | 4h | Retrieval-Augmented Generation on Azure Azure AI Search, embeddings, re-ranking, grounded citations |
| W3 | Saturday | 4h | LangGraph Deep Dive Graphs, cycles, subgraphs, streaming, human-in-the-loop interrupts |
| W3 | Sunday | 4h | Multi-Agent Systems & Guardrails Supervisor/worker agents, Content Safety, prompt-injection defenses |
| W4 | Saturday | 4h | Model Fine-Tuning & Evaluation on Foundry Dataset prep, fine-tuning jobs, side-by-side model evaluation |
| W4 | Sunday | 4h | Observability & Deployment Tracing, evals, Azure Container Apps deployment |
| W5 | Saturday | 4h | Microsoft Agent Framework Fundamentals Foundry-first Python SDK, agents, tools, MCP, agent sessions |
| W5 | Sunday | 4h | Agent Framework Workflows & Capstone Graph workflows, the Harness, LangGraph vs. Agent Framework, final capstone |
You'll deploy real models through Microsoft Foundry and provision real Azure services (AI Search, Cosmos DB, Container Apps) from Session 1. Azure credits are provided to each participant to cover this usage, so nothing in this program runs against a toy sandbox — you're learning on the same infrastructure you'll ship to.
Expand any weekend for the full breakdown. Each block builds directly on the one before it — by Weekend 5 you're comparing frameworks, not learning them from scratch.
Get a working agent talking to Microsoft Foundry, then make it reason on purpose.
Topics: LLM apps vs. agentic systems · ReAct, Plan-and-Execute & tool-use patterns · tour of the Microsoft Foundry model catalog, deployments & playground · Foundry SDK auth (Entra ID / keys) · wiring LangChain to Foundry-hosted models (GPT, Phi, Llama, Mistral)
Topics: Prompt patterns (zero/few-shot, CoT, ReAct, self-consistency) · structured output & function/tool-calling schemas · LangChain prompt templates & output parsers · context-window budgeting, chunking & context-assembly strategies · prompt versioning & testing
Give the agent a memory that survives a restart, and facts it can actually cite.
Topics: LangGraph state, threads & checkpointers · short-term vs. long-term memory · summarizing & entity memory · persisting agent state to Azure Cosmos DB / Azure Cache for Redis · multi-turn continuity
Topics: Chunking & embedding strategy · Azure AI Search (vector, hybrid & semantic ranking) · Foundry embedding models · advanced retrieval (re-ranking, HyDE, self-query) · grounding responses with citations
Turn one agent into a team, then put a safety layer around what it's allowed to do.
Topics: Graphs, nodes, edges & conditional routing · cycles & recursive reasoning · subgraphs & parallel branches · streaming · human-in-the-loop interrupts & approval gates
Topics: Supervisor/worker & agent-handoff patterns · connecting agents to tools, Azure Functions & MCP servers · Azure AI Content Safety · prompt-injection & jailbreak defenses · PII redaction & output validation · tool-permission policies
Customize the model, then watch the whole stack in production.
Topics: When to fine-tune vs. RAG vs. prompt-engineer · dataset prep for supervised fine-tuning · submitting & monitoring fine-tuning jobs in Microsoft Foundry · distillation basics · side-by-side model evaluation
Topics: Tracing with LangSmith / OpenTelemetry / Azure Monitor & Application Insights · cost, latency & token dashboards · automated regression evals & red-teaming · deploying agents to Azure Container Apps · Managed Identity & Key Vault for secrets
Learn Microsoft's own agent framework — see how the same agentic patterns you built in LangGraph translate to Microsoft's native, Foundry-first SDK, then finish with a capstone that ships to Azure.
Topics: Agents vs. workflows — when to use which · Agent Framework as the successor to Semantic Kernel & AutoGen · building agents with the Python agent-framework package against Microsoft Foundry via FoundryChatClient · tools & hosted MCP servers · agent sessions for multi-turn state · middleware for intercepting agent actions
Topics: Graph-based workflows — type-safe routing, checkpointing, human-in-the-loop · the Agent Framework Harness (planning & todo tracking, context compaction, file access/memory, tool-approval policies, built-in observability) · multi-agent orchestration patterns · migrating from Semantic Kernel / AutoGen
By the end of Weekend 5, you'll have a deployed, observable, guardrailed multi-agent system on Azure — built twice, in two frameworks — plus everything below.
Bring your own Python skills — leave with a production agentic stack on Azure, built with both LangGraph and Microsoft Agent Framework, and the Azure credits to keep building on it.
Submit an enquiry