#053 - Effective Use of Large Language Models in Engineering

A Guide to Practical Use Without Over-reliance

Hi all,

The holidays are fast approaching. Stay safe out there and enjoy it!

---

Let’s talk about the elephant in the room. Or rather the elephants in each engineering office around the world.

Large language models (LLMs) like ChatGPT, Claude, Gemini and Llama have become ubiquitous tools for enhancing productivity. Their potential to streamline specific tasks is obvious, but in engineering, their use demands deliberate care and discipline.

This article explores how LLMs can support engineers by organizing thoughts, iterating on ideas, and improving technical writing, while underscoring their limitations in performing calculations or design tasks.

Relying on LLMs for critical engineering decisions isn’t just misguided—it’s asking for trouble.

Their value isn't in replacing your thinking but in amplifying it.

A craftsman’s skill gives value to his tools, not the other way around.

There are still purists who scoff at using LLMs for tasks like writing, dismissing them as crutches. This mindset is reminiscent of the early resistance to calculators, with critics warning that reliance on them would erode mental math skills. While there’s a kernel of truth in that concern, calculators freed us to focus on higher-order problems by offloading routine tasks. The same principle applies here: by offloading aspects of drafting and refining, LLMs give engineers the mental bandwidth to deliver more value and tackle broader challenges.

Engineering is increasingly defined by how much you can achieve with the tools and time available. Those who refuse to adapt risk becoming like monks in a cloister, painstakingly crafting calligraphy while the rest of the world moves forward, solving old problems faster. Our industry, steeped in tradition for better or worse, must reckon with this reality.

I’ve seen this pattern play out before. My dad used to rail against smartphones, outraged by how younger generations spent their lives staring at screens. He wasn’t entirely wrong, anyone riding public transit can attest to humanity’s collective postural crisis, my own spine included. Yet, despite his disdain, he eventually capitulated. Now, with on-demand access to cattle prices, farming equipment reviews, and Irish horse racing results, you’d need a crowbar to pry his phone from his hands. Things change, and our industry is no different. Resistance fades when utility becomes undeniable.

It’s ultimately a personal choice, but the direction of progress is hard to ignore.

How I Use Large Language Models in My Engineering Work

I use large language models (LLMs) for a variety of tasks, but the most impactful applications in my day-to-day engineering work fall into three main categories: meeting transcription, ideation, and writing assistance.

1. Meeting Transcription and Note-Taking

One of the greatest challenges in detailed engineering meetings is balancing participation and note-taking. In technically demanding discussions, you can either contribute meaningfully or take notes—you can’t do both effectively, at least I can’t. I’ve set up a localized transcription tool, that uses my own hardware to record and transcribe my meetings without any third party services. This allows me to record the critical points of a conversation without sacrificing my ability to engage and frees me up to focus on adding value to the discussion while ensuring nothing important is missed.

2. Idea Iteration

I treat LLMs like a thought partner to help flesh out complex scenarios. For instance, when designing an intake in mountainous terrain, I can set the stage by describing the scenario in detail. Then, through a back-and-forth dialogue, I explore key topics such as hydrology, geomorphology, geology, geotechnics, structural design, mechanical equipment, and operational challenges. This iterative process sharpens my intent, organizes my thoughts, and refines ideas in a way that makes them actionable. Over time, the ideas become clearer, more focused, and easier to execute.

3. Writing Assistance

Whether it’s drafting reports, specifications, drawing notes, or even emails, LLMs are excellent writing assistants. Writing is situational; the tone, clarity, and intent must match the audience and purpose. LLMs help me craft messages that communicate technical concepts effectively while maintaining the right tone for the stakeholders involved. You do need to be very specific with your prompts or you get generic ‘AI’ outputs.

Beyond These Use Cases

While these are the most significant ways I use LLMs, I also lean on them for technical tasks like code generation, research and modelling help. I can discuss this more another day.

---

The Limits of Large Language Models in Engineering

LLMs are not engineering tools. They lack the ability to validate data, adhere to design codes, or understand the nuances of engineering standards - although this gap is closing. Using an LLM to perform calculations or handle design workflows invites errors because:

• LLMs process text and patterns, not numerical accuracy or real-world constraints.

• They cannot verify outputs against design standards like CSA, AISC, or Eurocode.

• Their responses are probabilistic and prone to misinterpretation of inputs.

In short, LLMs cannot and should not replace your expertise as an engineer. They are most effective when treated as a tool for structuring ideas, brainstorming, and refining your work, not performing it.

---

Prompt Engineering: Organizing Your Thoughts with LLMs

Interacting with LLMs effectively starts with asking precise, structured questions—a skill often called prompt engineering.

Much like writing a design brief, a clear prompt is essential. Avoid broad or vague requests like:

"Design a retaining wall."

Instead, provide the necessary context:

"List key design considerations for a cantilever retaining wall supporting a 5-meter high backfill with a sandy soil type. Focus on stability and drainage requirements."

This level of specificity helps the model focus on your actual needs and gets you started..

Break Down Problems into Steps

Complex engineering tasks often require multiple steps. Instead of asking for a complete solution, guide the LLM through discrete stages. For example:

1. Clarify concepts: "Explain the factors influencing soil bearing capacity in clay soils."

2. Explore options: "List methods for improving soil bearing capacity in poor conditions."

3. Refine focus: "Summarize key considerations for shallow foundations in clay soils."

This modular approach improves the accuracy and relevance of responses.

Iterate and Adjust

If the initial response isn’t quite right, refine the prompt. Treat this as a feedback loop:

• Add context: “Include considerations for frost depth in cold climates.”

• Focus scope: “Explain lateral earth pressure theories for active and passive conditions.”

Each iteration sharpens the output, much like refining a design or a code script.

---

LLMs as Writing Assistants: The Most Valuable Use Case

While LLMs should not be trusted with calculations, they excel as writing assistants. Clear communication is critical in engineering, whether drafting technical memos, project reports, or proposals. LLMs can help streamline your work, but the outputs must be critically reviewed. The reputational cost of an error when using AI tools is extremely high. Be diligent.

Start with Structured Data

The quality of LLM-generated text depends on the clarity of your input. Before using an LLM for writing, prepare a structured summary of your data or findings:

• Key points: Highlight major conclusions or insights.

• Data formats: Provide tables or bullet points where necessary.

• Intended audience: Specify if the writing is for clients, project teams, or internal documentation.

For example, instead of asking:

"Give me a template for a geotechnical report."

Provide specific inputs:

"Summarize the settlement analysis results for a proposed building on clay soil. Include a comparison of predicted settlements from empirical methods (per the attached csv) versus finite element modeling with baseline assumptions of ‘x’ as noted in report ‘y’. Ask questions to help clarify scope"

This ensures the LLM can produce a draft that aligns with your needs.

Draft, Then Refine

Ask the LLM to generate an outline before drafting full text. This keeps the structure logical and focused:

"Create an outline for a technical memo summarizing the results of wind load analysis for a 20-story high-rise. Sections should include: objectives, methods, results, and recommendations."

Once you have the outline, expand specific sections:

"Draft the methods section, describing the use of CFD simulations to analyze wind loads, including assumptions and boundary conditions."

Finally, refine the draft by asking the LLM to improve readability, tone, or conciseness:

"Polish this section to make it more concise and professional."

Verify and Edit

Always review LLM-generated content critically:

• Technical accuracy: Ensure data and conclusions are correct.

• Clarity: Adjust any convoluted or ambiguous language.

• Tone: Align the text with professional standards.

LLMs are tools, not autonomous authors. Your expertise ensures the final product meets the required standard.

---

Practical Applications for Python Workflows

For engineers using Python, LLMs can assist with code-related tasks without taking control of your work. Use them for:

• Refactoring code: "Simplify this Python function for calculating deflections in beams."

• Documentation: "Add docstrings to this script to explain its purpose and input parameters."

• Exploration: "Suggest libraries for automating geotechnical data visualization in Python."

While these tasks can save time, always test and validate code thoroughly. Treat the LLM as a collaborator, not a programmer. I've often observed junior engineers copying and pasting large blocks of code in and out of ChatGPT without understanding what’s happening—this is the very definition of spinning your wheels.

Slow down, zoom out and and break the problem or code into manageable, bite-sized chunks.

---

Common Pitfalls and How to Avoid Them

To use LLMs effectively, you must avoid certain traps:

• Over-reliance: Never delegate critical engineering tasks or decision-making to an LLM.

• Ambiguity: Vague prompts result in generic, unhelpful responses. Be specific.

• Uncritical acceptance: Always review outputs for accuracy, especially in technical writing.

LLMs are tools for improving efficiency, not substitutes for engineering judgment.

---

Conclusion: LLMs as Tools, Not Replacements

Large language models are valuable assistants when used appropriately. They excel at structuring ideas, iterating on concepts, and enhancing technical writing. However, their limitations make them unsuitable for performing engineering calculations or making design decisions.

Always remember that the responsibility for accuracy, clarity, and professionalism rests with you.

These tools will only get better, but becoming proficient requires practice and an understanding of their limitations and risks. I encourage all engineers to explore and experiment. As new advancements emerge, you'll be prepared to integrate them effectively into your professional practice.

Just double check outputs. We’ve all seen those whitepapers and technical publications where the author forgot to remove the ChatGPT generic text. Nobody wants that.

Be good and I’ll see you in the next one.

James 🌊