This course contains the use of artificial intelligence.

Build Four AI Agents That Watch a Live Industrial Plant, Diagnose Problems, and Recommend Action — Safely, With a Human Always in the Loop

Most agentic AI courses teach you to build agents that book flights or summarize emails. This course teaches you to build
agentic AI for manufacturing — where the data is real-time sensor data, the stakes are a live process, and an agent that
acts without oversight is not a feature. It is a risk.

You will build four agents, each one smarter than the last, connected to a real Unified Namespace running MQTT, Sparkplug B,
and OPC UA — the same protocols running on plant floors today.

Agent 1 — The Observer reads live plant data every 30 seconds and describes what is happening, in plain English.

Agent 2 — The Diagnostician does everything Agent 1 does, and when something looks wrong, reasons about why.

Agent 3 — The Advisor diagnoses the problem and recommends a specific action, with urgency and confidence attached.

Agent 4 — The Supervised Operator routes its own recommendation through a human approval gate before it acts — and logs every decision. Nothing happens in your plant without a human signing off, exactly the way Management of Change already works in your operation today.

What You Will Build

Using Python, LangGraph, and MCP (Model Context Protocol), you will build each of these four agents from scratch against a
realistic simulated manufacturing scenario — Walker Manufacturing, a polymerisation plant where a cooling pump failure develops slowly across a single night shift. By the end of the course, you will have working code for all four agents, plus a Streamlit oversight dashboard where a human approves every recommended action before it executes.

What You Will Learn

The architecture behind every production-grade industrial AI agent — observe, reason, recommend, act, and the human-approval
gate that makes the last step safe.

How to use LangGraph to structure an agent's reasoning as a graph, the same way ladder logic structures a PLC's program —
if you have ever read ladder logic, this will feel immediately familiar.

How to use MCP to give an LLM agent controlled, scoped access to a live industrial namespace — not generic API calls, but the
same tool-access pattern increasingly used in production agent architectures.

Why an agent needs ISA-95 context — site, area, line, equipment — to reason accurately, and what happens when it does not have that structure.

How to design a human-in-the-loop approval architecture that lets an agent recommend action in real time without ever making
an uncontrolled change to a live process — and why that design is not a limitation, but correct engineering.

How to build a full audit trail so every agent decision, every approval, and every action is logged and reviewable.

Who This Course Is For

This course is for IIoT and OT engineers who have built a Unified Namespace and now want to make it intelligent, not just
visible.

It is for software developers and AI practitioners who understand LLMs and agents generally, but have never connected
one to a real-time industrial system, and want to learn the specific architecture that makes that safe.

It is for maintenance, reliability, and process engineers who have relied on years of pattern recognition to catch problems
early, and want to understand how that judgment can be captured in a system that runs continuously.

It is for solution architects, presales engineers, and technical consultants in Industrial AI or Industry 4.0 roles who need
hands-on proof behind their architecture conversations, not just slide familiarity.

This course is not for you if you are looking for a general introduction to prompting ChatGPT, or a no-code workflow builder
course. Every lecture here builds real, runnable code against a live simulated industrial namespace.

Prerequisites

Basic Python is helpful but not required — every line of code is explained as it is written. Completing this instructor's UNS
Masterclass first is recommended, since this course builds on that namespace, but a Bonus section gets your stack running from
scratch if you are new. Total LLM API cost to complete every exercise is typically under five dollars.

Hands-on labs. Lifetime access.

More Info