Course outline
Level 1 in detail: what the curriculum covers
Each module connects practical AI capability to worker safety, qualified review, intellectual-property protection, cybersecurity, authorized system access, approved procedures, manufacturer instructions, auditability, escalation, and accountable decisions.
Modern Artificial Intelligence in a Manufacturing Company
What the systems sold to a plant as "AI" actually are, and why that changes how you check them. This module separates fixed automation, rules, statistical process control, machine learning and generative AI, explains how a language model builds an answer, opens up the retrieval, tools and agents inside plant products, and maps the categories a small manufacturer meets.
-
Fixed Automation, Rules, Machine Learning and Generative AI
35 min
-
How a Language Model Builds an Answer
40 min
-
Retrieval, Tools and Agents Inside the Products You Are Sold
40 min
-
A Field Guide to Plant AI Categories and the Public Help You Can Reach
45 min
What AI Can and Cannot Reliably Do With Plant Information
Fluent output is not checked output. This module shows how a tool invents a standard, a revision or a part number, how tolerance and unit arithmetic fails silently, how to read an accuracy figure next to the recall it is hiding, and why the strongest countermeasure to over-trust is a habit rather than a warning label.
-
Fabricated Standards, Revisions and References
35 min
-
Tolerances, Units and the Limits of Context
40 min
-
Reading Accuracy, Recall and F1 Honestly
45 min
-
Automation Bias and the Discipline of Doubt
40 min
Export-Controlled Technical Data, Deemed Exports and AI Tools
The sharpest AI trap in manufacturing. This module explains what technical data is, why releasing it to a foreign person inside the United States can itself be an export, what the published-information and end-to-end-encryption provisions really cover, who must register, and how to decide whether a drawing may enter a system at all.
-
What Counts as Controlled Technical Data
40 min
-
The Deemed Export Rule and What a Tool Does With a File
45 min
-
Published Information and Encryption: What They Do Not Cover
40 min
-
Screening a File Before It Reaches Any System
45 min
Trade Secrets, Process Knowledge and What Never Leaves the Plant
Your parameters, your heat-treat recipe, your yields and your cost model are the plant's crown jewels, and federal law protects them only while you keep them secret. This module covers the statutory definition, the reasonable-measures element a pasted prompt can quietly weaken, what a federal civil action involves, the notice employers owe, and how to classify plant data.
-
What Makes Process Knowledge a Trade Secret
35 min
-
Reasonable Measures and the Outside Tool
45 min
-
When a Secret Walks: Misappropriation, Remedies and the Notice Employers Owe
40 min
-
Classifying Plant Data Before a Tool Sees It
40 min
Who Owns the Decision: Engineering, Quality, Safety and Production Accountability
A tool can draft, extract, summarise and flag. It never dispositions a part, releases a lot, seals a drawing or clears a machine. This module covers the duties that do not move when AI helps: the employer's duty under the general duty clause, the named quality authority, the state question of whether a document needs a seal, and a code that binds only its members.
-
The Employer Duty That Does Not Move
35 min
-
The Named Quality Authority and What AI Never Signs
40 min
-
Does This Engineering Document Need a Seal? Two State Examples
40 min
-
A Professional Code Binds Its Members, Not the Plant
35 min
Law, Standard, Certification Scheme or Contract: Telling Them Apart
An OSHA rule is law. A quality standard is a document you buy. Certification is a scheme sold by a registrar or demanded by a customer. And an automotive or aerospace requirement binds you through a purchase agreement, not a statute. This module teaches the four apart, places the quality standards accurately, and shows where the voluntary AI management standard fits.
-
Four Kinds of Obligation in a Plant
40 min
-
Where the Quality Standards Actually Sit
40 min
-
A Customer Requirement Is a Contract, Not a Law
40 min
-
The AI Management Standard and What a Vendor Claim Is Worth
35 min
OSHA Duties and the AI Standard That Does Not Exist
There is no OSHA artificial intelligence standard, directive or guidance document. This module teaches the duties that do bind a manufacturing employer — the general duty clause, machine guarding, the control of hazardous energy, personal protective equipment, hazard communication and injury recordkeeping — and shows how an AI-related hazard is reached through them.
-
The Employer Duty and the OSHA AI Standard That Does Not Exist
35 min
-
Machine Guarding and Hazardous Energy When Equipment Gets Smarter
40 min
-
PPE Hazard Assessments and Chemical Information Drafts
35 min
-
Injury Recordkeeping, State Plans and What National Figures Can Say
35 min
Robots, Cobots and Machine Safety When Perception Is Learned
What changes when a stop, a speed limit or a separation distance depends on a model that can be wrong. This module separates a safety function from a perception feature, works through the robot-system failure modes OSHA itself documents, names the robot and machinery safety standards with their status and edition, and explains why a vision system is never the safeguard.
-
A Safety Function Is Not a Perception Feature
40 min
-
How Robot Systems Actually Hurt People
40 min
-
Robot Safety Standards, and Which Edition Applies
40 min
-
Collaborative Applications, Exoskeletons and the Integration Contract
40 min
Prompt Engineering for Manufacturing Professionals
How to ask for plant work in a way you can check: the data screen that happens before you type, a nine-part prompt built around a supplied released document rather than a recalled one, disciplined iteration, and patterns for work instructions, nonconformance narratives, maintenance summaries and supplier email. Every output is a draft until a named person owns it.
-
The Screen That Happens Before the Prompt
35 min
-
A Structured Prompt for Plant Work
40 min
-
Iterating Without Drifting From the Source
35 min
-
Prompt Patterns for Plant Documents and Supplier Correspondence
40 min
Verifying a Standard, a Regulation and a Revision Level
An assistant that names a standard, a regulation or an edition is giving you a lead to check, not a citation to quote. This module builds a research workflow that confirms a reference exists, is current, and says what is claimed for this product and this customer, using a withdrawn quality standard edition and a superseded medical-device provision as the worked examples.
-
The Assistant Is Not the Authority
35 min
-
Withdrawn, Superseded or In Revision: Reading a Status Field
40 min
-
Does This Requirement Reach My Part, My Plant and My State?
40 min
-
The Verification Record That Survives an Audit
35 min
Drafting Work Instructions, Quality Records and Supplier Communication
How to get a first draft without a first mistake. This module covers drafting work instructions, maintenance procedures, nonconformance narratives and supplier correspondence from a verified set of released documents, what the record must show about preparer, reviewer and evidence, and why no AI tool drafts a disposition, a release or a conformance statement.
-
First Draft Without First Mistake: Working From a Verified Set
40 min
-
Where Drafting Stops: Safety Content in Instructions and Procedures
40 min
-
Nonconformance Narratives and Corrective Action: Drafted, Never Dispositioned
40 min
-
Electronic Records, Audit Trails and What You Send a Customer
40 min
Machine Vision Inspection and Predictive Maintenance: What the Evidence Actually Shows
Read a detection claim like an engineer. This module takes apart accuracy, precision, recall, F1 and mean average precision, shows what the most-used public datasets actually contain, converts a published miss rate into escaped defects on your own volume, and sets out what a pilot on your own material has to look like before anyone believes a number.
-
Accuracy, Precision, Recall, F1 and mAP: What Each One Hides
45 min
-
What the Public Datasets Actually Are
40 min
-
From a Miss Rate to Escaped Defects on Your Own Volume
45 min
-
A Pilot on Your Own Material, and the Trust Problem Behind It
45 min
Building Repeatable Plant AI Workflows
How a one-off chat becomes a controlled plant process: a six-stage workflow canvas with a trigger, a data-class screen, a named AI step, verification, a named approval and a record. It also covers designing so a wrong output cannot become a wrong action, where an agent's authority must stop, and how to keep the review step honest.
-
From a Chat Window to a Controlled Plant Workflow
40 min
-
Designing So a Wrong Output Cannot Become a Wrong Action
40 min
-
Agents in the Plant and Where Their Authority Stops
40 min
-
Keeping the Review Step Honest and Recording the Approval
40 min
Plant Cybersecurity, Vendor Claims and the OT Boundary
Where an AI tool may sit in a plant and where it may not. This module draws the operational-technology boundary, sets the asset inventory and access basics a plant without security staff can run, teaches you to read a vendor's data-handling claims for what they avoid saying, and shows how a federal contract clause turns a recommendation into an obligation.
-
The OT Boundary and Why No AI Tool in This Level Crosses It
40 min
-
Asset Inventory and Access Basics Without a Security Team
40 min
-
Reading a Vendor's Security and Data-Handling Claims
45 min
-
When a Federal Contract Clause Sets Your Security Floor
40 min
Practical Manufacturing AI Lab
Four hands-on labs on fictional plants. Find the planted errors in an AI-drafted work instruction and nonconformance report; catch a fabricated standard citation and a wrong drawing revision, then rebuild the weak prompt; screen a quote package for controlled technical data and trade secrets; and take apart a vision-inspection claim before drafting a plant AI use policy.
-
Lab One: The AI-Drafted Work Instruction and Nonconformance Report
45 min
-
Lab Two: The Fabricated Citation and the Wrong Revision
45 min
-
Lab Three: Screening a Quote Package and Designing the Workflow
45 min
-
Lab Four: The Vendor Claim and the Plant AI Use Policy
45 min
Capstone: The Plant AI Implementation Plan
The plan you will actually write: choose and scope one low-risk problem, classify every data class it touches and set the controls, name the duties that do not move to a tool, ask the vendor the questions that matter, and define the baseline your own plant measures, the failure indicators, the stop conditions and the improvement loop.
-
Choosing and Scoping a Low-Risk First Problem
40 min
-
The Data Classes in Your Plan and the Controls They Require
40 min
-
Duties That Do Not Move, the Vendor Question and the Documentation
40 min
-
The Baseline, the Stop Conditions and the Improvement Loop
45 min