AI for Developers

Advanced Agents with LangGraph

3 - 6 Hours 4 Modules

Overview

Main Theme: AI for Developers

Duration: 3 - 6 Hours

Level: Advanced / Engineering

Target Audience: Software Architects, AI Engineers, and Senior Developers

ABOUT THE TRAINING

LangGraph enables AI to move beyond just chat and operate as a robust state machine. This training teaches how to build applications that don’t "get lost" in the conversation, allowing for the creation of loops, complex logical conditions, and memory that persists between sessions. It is the ideal tool for mission-critical processes where reliability and auditing are essential.

Objective

Empower the professional to design and implement agent flows that can fail, correct their own error, and try again, ensuring that each decision is traceable and that sensitive actions go through human approval.

TRAINING MODULES

State Machines and Cyclical Flows

Learning about graph architecture for AI. The student will learn to create cyclical flows where the agent can attempt to execute a task, identify a failure, and perform self-correction autonomously before delivering the result.

Persistent Memory and Long-Term State

Implementation of data persistence. Unlike common chats, this module focuses on how to make the agent recall interactions from days or weeks ago, maintaining the historical context of decisions and user preferences in a structured manner.

Human-in-the-loop (Human Approval)

Development of checkpoints. The student will learn to set up "break points" where the agent pauses its execution and waits for a human to validate or edit an action before it is completed.

Case Study: Multi-Agent Orchestration "Ask David"

The challenge is to architect a "Maestro" (Supervisor) that delegates tasks to specialized agents: a structured data agent (Text-to-SQL), a document agent (RAG), and an analytical agent, integrating everything into a single high-accuracy decision flow.

A case showing LangGraph in use

J.P. Morgan, a global leader in financial services, built a robust AI system.

Ask David is a real, advanced example of how large companies are moving AI from simple chatbots to agentic systems.

The name DAVID stands for: Data, Analytics, Visualization, Insights, Decision-making.

It was built for the bankers at JP Morgan Private Bank. Picture an advisor deciding whether to recommend an investment fund to a billionaire client: they cannot get it wrong. "Ask David" automates research that used to take hours, cross-referencing spreadsheet data (structured) with research reports and news (unstructured).

Architecture built with LangGraph

Main agent:

Supervisor agent (the conductor): It reads the question first. It does not try to answer on its own: it analyzes the user's intent and decides which specialized agent to call. If the question is complex, it builds an execution plan, calling several agents in order.

Sub-agents:

Structured data agent (Text-to-SQL): Its only job is talking to databases. It turns "What was the return of fund X in 2024?" into a real SQL query, making sure the number that comes back is exact and from the official source.
Unstructured data agent (RAG): It focuses on text. It reads PDFs, meeting transcripts and emails. Ask "Why was this fund closed?" and it searches the research report and summarizes the reason.
Analytical agent (ReAct): It uses internal tools and APIs to run mathematical calculations or financial simulations in real time.

Source: zenml.io