It’s not the algorithm. It’s everything around it.
Jesse Keckler
Biomedical Engineer, Nottingham Spirk
Artificial intelligence is reshaping industries, from automating complex manufacturing systems to pushing the limits of what’s possible in medical devices. But turning AI’s potential into effective, real-world implementations is an entirely different challenge.
As a biomedical engineer, I’ve spent my career translating human behavior into data that drives safer, smarter systems. In this post, I’ll take you inside a motion-tracking AI project I led in a manufacturing environment, one aimed at improving worker safety and operational efficiency. It’s not a story about hype. It’s about the real work: gathering usable data, adapting to unpredictable conditions, and learning to make AI deliver in the field.
Starting with the right questions
Launching any AI initiative is like planning an expedition: you’re not going far without a good map. During early research, tools like Perplexity were invaluable, helping us surface insights quickly and, more importantly, ask the right questions.
In our case, that meant identifying which worker motions were most injury-prone. We dove deep into the biomechanics of repetitive strain, prioritizing high-risk movements and laying the foundation for an AI system focused on ergonomics and injury prevention.
From plan to practice
On paper, our task seemed straightforward: collect motion data via 9-axis IMUs, label it, train the model, and deploy it. In practice, it was anything but.
Labeling quickly became a bottleneck. Even with synced video, identifying motions from IMU (Inertial Measurement Unit) data proved difficult. Manually matching sensor readings to physical actions was both time-consuming and error-prone.
The real challenge came when we finally received data from actual factory workers late in the project. Their movements differed significantly from our proxy users: they were faster, more variable, and less predictable. Unsurprisingly, the model struggled to adapt.
Hardware realities
Hardware friction added to the complexity. We had planned for real-time IMU streaming, but frequent disconnects forced us to switch to recorded data, complicating processing and slowing iteration.
Still, we delivered a strong proof of concept. While the system fell short of our original performance goals, it provided real insight into how motion data could enhance both safety and operations. More importantly, it reinforced a hard truth about AI: success is iterative. Across the model, the data, and the development pipeline, we had to constantly reassess. We trained on four separate datasets, each one exposing new gaps and forcing us to rethink our approach.
What I would do differently
If I could do it again, I’d push much harder to get access to real-world data from day one. Having the right data upfront would have saved us months and likely led to a far more accurate model.
The real breakthrough came during a visit to the plant. Watching the work unfold in real-time, we realized this wasn’t just about tracking motion. It was about understanding the flow of people, tools, and parts within a dynamic system. We saw how workers adapted on the fly, how roles shifted in response to delays, and how inefficiencies quietly emerged. Suddenly, AI wasn’t just about pattern recognition—it became a tool for identifying bottlenecks, surfacing opportunities, and even recognizing human creativity in the system.
The bigger picture
This project highlighted a broader challenge in AI adoption: your model is only as good as your data. And collecting quality, representative data at scale remains one of the biggest hurdles.
In future projects, I plan to explore semi-supervised learning approaches to reduce manual labeling and improve scalability. But one thing is clear: despite the friction, the potential is real.
AI has the power to transform how we design systems, solve operational problems, and support human adaptability in real-time. But to get there, we need to approach these projects with clear eyes, patience, and a willingness to lay the groundwork the right way.
The promise is real—but only if we build the foundation to support it.
About Nottingham Spirk
Nottingham Spirk is a world-class product innovation firm with an unrivaled record of developing and commercializing disruptive consumer products, medical devices, digital IoT products and connected industrial products. We collaborate with Fortune 1,000 companies, middle market companies and funded venture companies to discover, design and execute product innovation programs and strategic business platforms that will delight customers, grow markets, and generate new revenue streams. Learn more about our AI-enabled hardware and firmware solutions.