Did you know that Generative AI (GenAI) can help engineers diagnose car issues in seconds? Or predict equipment failures before they happen? GenAI is making these possibilities a reality by speeding up data analysis and algorithm development, enabling engineers to apply their domain expertise and uncover actionable insights.
Engineering teams generate terabytes of data every year, with Gartner estimating that up to 80% of this information is unstructured. Service records, research papers, and technician notes contain critical institutional knowledge, but their inconsistent formats make them difficult to parse. GenAI tools help engineers combine structured and unstructured data, enabling large-scale analysis that was previously impractical for many engineering teams. For engineers, this means quicker troubleshooting, improved design workflows, and accelerated discovery.
Engineers’ GenAI Blind Spots
Despite GenAI’s versatility in reshaping engineering work, there remains a gap between what the technology can enable and how it’s being used day-to-day. Many engineers still primarily use GenAI for low-level code generation or documentation, rather than for more advanced workflows.
To better understand engineers’ attitudes toward GenAI and its real-world applications, MathWorks conducted an informal social media poll in December 2025. The responses revealed several insights, including:
• 83% of engineers use GenAI at least once a month. The most common GenAI use cases among respondents were “Writing Code” and “Documentation and Reporting.”
• The top concern among engineers related to GenAI is its integration into existing workflows (46%). This number rose to 75% among engineers with at least six years of experience.
This data indicates that the majority of engineers are using GenAI, but not for strategic engineering tasks. Engineers seeking to expand their GenAI knowledge should consider using it to prepare and analyze unstructured data.
Using Service Manuals, Engineering Documents, and Mechanic Reports to Create a Chatbot for Technicians
Automotive troubleshooting requires diagnosing complex issues across diverse vehicle makes and models. Although large language models (LLMs) contain extensive public automotive knowledge, they lack detailed, brand-specific data. To bridge this gap, engineers at Tata Motors applied a GenAI technique called Retrieval-Augmented Generation (RAG), which combines the broad knowledge of LLMs with proprietary data to generate context-specific guidance.
The engineers used RAG to develop a context-aware chatbot that retrieves internal documents and uses them to generate troubleshooting responses. MATLAB® was used to develop a RAG workflow, enabling their app, known as ServiceSage, to search service manuals, engineering documents, and mechanic reports. When a mechanic poses a question to ServiceSage, it is converted into a numeric representation that the GenAI tool can understand, and the system finds the most relevant documents. The wording of the question is largely irrelevant because RAG performs a semantic search, making inferences based on related concepts. These documents are then fed to the AI model, which combines them with its general knowledge to produce a clear, human-readable answer.
This method is cost-effective and scalable, eliminating the need for expensive model retraining and handling large volumes of previously underused text data. The approach enabled the team to rapidly identify root causes, deliver context-specific guidance, and decrease repair turnaround times. With GenAI, engineers analyzed large volumes of text data efficiently and incorporated it into their troubleshooting workflows.
Tapping Global and Historical Scientific Research to Drive Food Science Discovery
Scientific research often spans thousands of papers published over decades and across various regions, making it impractical to examine all the studies on a given topic or to uncover hidden relationships without advanced analytical tools. Food science researchers at the University of Copenhagen faced this challenge when analyzing a vast amount of material to find connections between topics. LLMs can summarize individual documents but struggle to identify links across massive datasets. To overcome this, the researchers combined GenAI with classical techniques—text preprocessing and cleaning and information extraction—to impose structure on unstructured text before applying LLMs.
The University of Copenhagen team used GenAI in multiple steps throughout this process, including:
1. Cleaned and standardized thousands of PDFs with inconsistent formatting.
2. Generated missing keywords where metadata was lacking.
3. Converted the text into tokens and flagged unusually long words to identify buried chemical names.
4. Created a knowledge graph after breaking research papers into paragraphs and keywords. Each node in the graph represented a paragraph or chemical name, and the connections between nodes captured topic relationships.
The team then used MATLAB to apply graph theory to this dataset, identifying paths that linked concepts. They then processed these structured subsets using an LLM, which generated summaries and explained relationships that could take humans weeks to trace. The result was a workflow that transformed siloed research into actionable insights, accelerating discovery in food science.
Despite GenAI’s value – researchers saved themselves days’ worth of manual tasks using this process – the researchers’ success hinged on human judgment and manual work. They invested hundreds of hours running experiments and preparing data before they could upload it to GenAI. Discovering that paragraphs provided optimal text segmentation required repeated trial and error, as GenAI could not make this decision automatically. GenAI is powerful only when paired with proper data and methodical engineering.
Turning Maintenance Data into Foresight
Predictive maintenance (PdM) has traditionally relied on numerical sensor data, tracking everything from thermal changes to vibration and pressure, to detect patterns that precede equipment failure. Many organizations also collect valuable text-based information like maintenance logs and technician notes. These sources capture context that sensors alone cannot, revealing symptoms, documenting repair actions, and identifying suspected root causes.
This text is not a replacement for sensor data; engineers can use GenAI to make it more accessible and usable alongside traditional signals. For example, GenAI can help summarize maintenance histories, standardize inconsistent terminology, or label events such as component failures or recurring fault types. These labeled outcomes can then be aligned with time-series sensor data, providing clearer targets and context for PdM model development.
GenAI can also support the engineering workflow. Engineers can use it to draft and refine code for data cleaning, feature engineering, or exploratory analysis, or to evaluate alternative modeling approaches. However, domain expertise remains critical at every step. Only experienced engineers can judge whether features are physically meaningful, whether model behavior aligns with known system dynamics, and whether outputs reflect real failure modes or simply artifacts of the data.
As with any PdM approach, GenAI-assisted workflows require careful validation before deployment. A model that performs well in a prototype or limited test set may not generalize to real-world operating conditions. Engineers should validate results using representative datasets, stress-test models across operating regimes, and apply deterministic checks to ensure robustness. These best practices apply equally to GenAI and non-GenAI methods, reinforcing the principle that successful PdM depends not just on advanced tools, but on sound engineering judgment.
Engineers are Scratching GenAI’s Surface
Like any tool, GenAI should be used methodically rather than applied to every problem. GenAI is most valuable when problems involve large volumes of unstructured data or require flexibility in interpreting language-based inputs. Engineers must integrate it strategically, expand their domain expertise, and examine how to apply this tool in their design methodology.
