GENAI FOR ENGINEERS
Generative AI for Engineers: What LLMs Can and Cannot Do for Technical Work
An LLM predicts the next token; that explains both its fluency and its invented part numbers. What engineers can safely use it for, and how to learn it free.
Generative AI for engineers means using large language models (LLMs) for technical work: drafting documents from data, searching manuals, explaining code and automating paperwork, while knowing exactly where they fail. An LLM predicts likely text, so it is fluent but not reliable on numbers, part codes or your plant. You can learn to use it properly, free, on edwartens.com.
This lesson by IBM Technology sits in the second module of the free Generative AI and LLM Foundations course, "How an LLM works: tokens in, probabilities out". The module's aims are to explain next-token prediction and why fluency is not evidence, to describe the tokeniser, embeddings, attention and parameters in one line each, and to predict which questions a model will be weak at. Its practice task counts the tokens in a page of a drive manual, then asks a free model to add up power readings and checks the answer in Python.
How an LLM works, in five lines
- Tokeniser. Text is split into tokens: whole words, pieces of words, digits and punctuation.
- Embeddings. Each token becomes a list of numbers that places it near tokens with similar meaning.
- Attention. Each position in the text weighs every earlier position to decide what is relevant. The idea comes from the 2017 transformer paper, Attention Is All You Need.
- Parameters. Billions of learned weights, fitted during training on a very large body of text.
- Output. A probability for every possible next token. The model picks one, appends it and repeats.
That last line explains most of what surprises engineers. The model is not looking anything up. It is producing the most plausible continuation. A made-up flange rating or a drive parameter number that does not exist is not a malfunction; it is the model doing what it was built to do, without the facts to anchor it. It also knows nothing after its training cut-off, and nothing about your plant unless you give it the text.
Generative AI for engineers: where it helps
Used well, an LLM saves real time on work that is mostly language:
- Documents from data. Turning an I/O list into description text, a test record into a FAT report section, or notes into a procedure draft, with a rule that anything not in the data is left blank.
- Searching manuals. With retrieval-augmented generation (RAG), the model answers from the pages you index and cites them.
- Explaining and drafting code. Python for data work, structured text boilerplate, spreadsheet formulas. Always run and review. AI copilots for PLC programming compares what Siemens, Rockwell, CODESYS and Beckhoff now build into their tools.
- Summarising. Shift logs, incident notes, long email threads.
- Workflow automation. An agent that reads a form, looks something up and drafts a reply for a person to approve.
Where it fails
- Arithmetic and unit conversions. Tokens are not numbers. Ask for the calculation as code, then run the code.
- Part numbers, parameter numbers, ratings. Check every one at the source.
- Your plant's specifics. It has never seen your P&IDs, interlocks or alarm philosophy.
- Safety functions. Never let a model design, change or approve anything in a safety instrumented system or a machine safety circuit.
Confidentiality matters as much as accuracy. Company drawings, customer data and network details should not go into a public chatbot. Running an open model on your own laptop with Ollama is one way to keep data local; the foundations course shows how. For organisations, NIST's AI Risk Management Framework is a widely used reference, and in Europe the EU AI Act deadlines apply to some industrial uses.
Prompting, RAG or fine-tuning
Most engineering needs are solved by the first two. A good prompt has five parts: role, task, context, output format and constraints, including a refusal rule such as "if the value is not in the data, write NOT FOUND". That single rule turns silent invention into a visible gap.
When the answers must come from your own documents, use RAG: split the manuals into chunks, embed them, retrieve the most relevant chunks for each question, and have the model answer only from those with page citations. Fine-tuning changes the model itself and is rarely the first answer for engineering teams, because it does not keep the model up to date with changing documents.
Agents, with a person in the loop
An agent is an LLM that can call tools: search a database, read a file, send a message. That is where most of the risk is, because the model's output now causes actions. The safe pattern is simple: give the agent the fewest tools it needs, make it read before it writes, and put a human approval step in front of every write or send. The OWASP Top 10 for LLM Applications lists "excessive agency" among its 2025 risks for exactly this reason. If you are building one near company systems, the free AI Security and the OWASP Top 10 for LLMs course covers prompt injection and least privilege.
A free route
- [Generative AI and LLM Foundations](/free/generative-ai-and-llm-foundations), beginner. Tokens, context windows and hallucination; choosing hosted or local models; calling free models from Python through OpenRouter; running open models with Ollama; embeddings, RAG, function calling and agents; and a datasheet assistant tested against a real question set.
- [Prompt Engineering for Engineers](/free/prompt-engineering-for-engineers), beginner. Structured prompts, few-shot examples, JSON output, and engineering documents generated from tables without invented facts.
- [RAG and Chatbots with LangChain](/free/rag-and-chatbots-with-langchain), intermediate. A manual-reading assistant with hybrid search, citations, refusals and a proper evaluation.
- [AI Agents and Workflow Automation with n8n](/free/ai-agents-and-automation-with-n8n), beginner. Workflows and agents on free or local models, with human approval gates.
The first, second and fourth are grouped on the AI foundations path. To compare these with other providers' free courses, read Free Generative AI and Prompt Engineering Courses With Certificate, Compared. For a worked build, see Build a RAG Chatbot for Equipment Manuals, and for the wider language toolkit read natural language processing for engineers.
Start the free course
Create a free account and start with Generative AI and LLM Foundations. No paid account is needed anywhere in the course. It ends with one 15-question final assessment, a 60% pass mark and three attempts, then a 24-hour wait and a fresh paper.
Learning is free in full. The optional EDWartens Certificate of Completion is a small one-off fee, US$8.99 for a beginner course and a little more for an intermediate one such as RAG. Anyone can check it at edwartens.com/verification. It is not a certification from any AI company and is not an accredited qualification.
Take the free course
FreeAI and machine learning · Beginner · Free
Generative AI and LLM Foundations
FreeAI and machine learning · Beginner · Free
Prompt Engineering for Engineers
FreeAI and machine learning · Intermediate · Free
RAG and Chatbots with LangChain
FreeAI and machine learning · Beginner · Free
AI Agents and Workflow Automation with n8n
Questions
What is generative AI in simple words?
Generative AI is software that produces new text, images or code from patterns learned in large amounts of data. Large language models are the text kind: they predict the next token again and again.
Why do LLMs make up part numbers and specifications?
Because they generate the most likely continuation, not a retrieved fact. A plausible-looking part number is exactly what the model is built to produce, so any figure it gives must be checked against the source document.
Can an LLM write PLC code?
It can draft structured text or describe ladder logic, and it can be useful for boilerplate. Every line must be reviewed and tested in a simulator, because a model has no knowledge of your machine, interlocks or safety functions.
Do I need to code to learn generative AI?
Not to start. Eight of the twelve modules in the free Generative AI and LLM Foundations course need no coding; the API and project modules use basic Python.
Is the generative AI course free?
Yes, in full, and it uses free models only. The optional EDWartens certificate for this beginner course is a small one-off fee, shown at checkout.
Sources
- Vaswani et al. (arXiv): Attention Is All You Need
- OWASP Gen AI Security Project: Top 10 for LLM Applications (2025)
- NIST: AI Risk Management Framework
- Ollama: run open models locally
Written by the EDWartens engineering team for general education. Product names are trademarks of their owners; mentioning them does not imply endorsement. Prices and terms of other providers were checked on the date shown and can change.
Free Machine Learning Course for Mechanical and Electrical Engineers
Machine Learning for Beginners: What It Is, How It Works and How to Start Free
Deep Learning Explained: Neural Networks, CNNs and LSTMs, and When You Need Them
Computer Vision for Industrial Inspection: How It Works and How to Learn It Free
TensorFlow vs PyTorch for Engineers: Which Should You Learn First?
Best Free AI Courses With Certificate for Engineers (2026)