Engineers and designers of technical systems often look to existing systems for inspiration – whether it’s a cooling device, a valve, or a piece of machinery. But understanding how these systems work can be a challenge, especially when the information is buried in lengthy technical documents and contains cross-domain knowledge. What if AI could help with that?

Recent advances in large language models (LLMs) are making it possible to automatically generate clear, structured summaries of how systems function. In fact, the author has participated in  developing  an AI tool for this purpose. The tool uses LLMs to explore online information about a device and translate it into a simplified explanation of how it works. This can save designers valuable time and help them discover useful ideas from a wide range of existing artifacts.

In this article, I will share how the tool works, what it can do, and how it might help make technical knowledge more accessible for creative problem solving.

Understanding SAPPhIRE: A Designer’s Lens on How Systems Work

Before we dive into how AI helps, it’s worth mentioning an important idea that underlies it - the SAPPhIRE model of causality (Chakrabarti et al., 2005).SAPPhIRE is a way to describe how a system works – not just what it does, but how it does it. It breaks down a system into seven layers: 

• States

• Actions

• Parts

• Phenomena

• Inputs

• oRgans

• Effects 

While this may sound complicated, dividing a system into these layers actually simplifies it.

Let’s take the example of a thermoelectric cooler which is an electric component that you can find in everyday devices such as air conditioners. The system includes Parts like semiconductor modules and metal plates, along with surrounding air or a heat sink. These Parts have specific properties and are arranged in a way that enables heat transfer—this constitutes the oRgans. When a voltage (Input) is applied across the terminals, it activates the Peltier Effect, a principle that governs how electrical energy can move heat across materials. This triggers a physical Phenomenon—heat starts flowing from one side of the cooler to the other, creating a temperature difference. As a result, the State of the system changes: one side becomes cold, the other hot. At a higher level, this results in the observable Action of cooling.

In short, breaking a system down using SAPPhIRE layers gives  designers a map that helps to see the full picture – from the nuts and bolts to the underlying physics. This kind of structured understanding is especially useful when looking for inspiration across different domains, like borrowing ideas from a cooling system to design a new kind of wearable device.Traditionally, building such a model takes time and expertise. That’s where our AI tool comes in – it helps automate this process, making it faster and more accessible.

Leveraging Large Language Models

We developed a tool that uses recent advances in large language models (LLMs) to automatically generate SAPPhIRE-style explanations of how a system works (Majumder et al., 2025). The tool searches for information about a device – like a solenoid valve or a thermoelectric cooler – and then organizes that information into the seven SAPPhIRE abstraction levels.

But it doesn’t just summarize. It reasons, checks, and refines. Using a method called Retrieval-Augmented Generation (RAG), the tool first gathers information from external sources—such as Wikipedia or any other technical documents available online or offline. It then drafts a structured explanation and cross-checks it against the retrieved content to ensure that the output is both accurate and grounded in real-world knowledge. This process helps designers quickly understand unfamiliar systems, opening the door to creative analogies and new ideas.

Where to Learn More

This article is based on our research paper titled "Development and Evaluation of a Retrieval-Augmented Generation Tool for Creating SAPPhIRE Models of Artificial Systems," presented at the 10th International Conference on Research Into Design (ICoRD 2025). The paper details how the tool was developed, the methods used, and the results from testing it on real-world systems. To explore the technical details or cite the work, you can access the full paper here https://arxiv.org/abs/2406.19493.