Makoto: Real-Time Computer Vision That Turns Production Line Observation into Data-Driven Business Improvement

Brief

Makoto helps industrial clients improve their production operations, but before they could recommend improvements, they needed data, and getting that data meant sending people onto the factory floor with stopwatches and clipboards for multiple days at a time. We built a fully on-premises computer vision system that does that work automatically. Multiple cameras track components moving through the workshop in real time, a custom-trained object detection model identifies what’s being tracked, and a web dashboard gives Makoto live and historical visibility into exactly where units are, how long they spend in each section, and where bottlenecks are forming.

Partnering with AIA has been an exciting journey to improve our data capture. We had a feeling that computer vision and machine learning would help us improve the accuracy and insights of our traditional data capture in environments where sensors and machine logged data does not tell the full picture. This solution has supercharged our efforts with the client.

Daniel, Consulting Partner at Makoto Asia Pacific

Problem

Makoto’s value to clients is their ability to identify where production processes slow down and to demonstrate, with evidence, that the changes they recommend actually work. Both halves of that depend on data. The problem was how that data was collected. Monitoring a production line manually means putting people in the workshop for days at a time, timing each task, tracking component movement by hand, and recording everything consistently enough to be useful. That process was labour-intensive, costly for clients, and vulnerable to the inconsistencies that come with any manual observation work. Worse, once the initial observation was complete and changes were implemented, going back to measure improvement meant repeating the entire manual process, which was rarely feasible given the cost, leaving Makoto with no reliable way to prove to clients that the changes they’d recommended had actually moved the needle.

Solution

We designed and deployed a fully on-premises computer vision system across Makoto’s client workshop. Multiple cameras were installed to cover the production floor, giving the system visibility across all key stages of the line. We trained a custom object detection model to recognise the specific components being tracked, ensuring accurate identification across different lighting conditions, angles, and production scenarios.

Figure 1: Object detection on production line components.

To protect client data and keep everything contained, we deployed the system on an edge computer installed on-site, so no data leaves the workshop, no cloud dependency, no privacy risk for the client’s operations. On top of the detection and tracking infrastructure, we built a web-based dashboard that gives Makoto both live and historical views of the production floor.

Video 1: Live computer vision in action.

• Custom-trained model: the object detection system was trained specifically on Makoto’s client components, not a generic model applied to an unfamiliar context
• Fully on-premises: edge computer deployment means zero data leaves the workshop, complete privacy for the client’s operations
• Live and historical data: the dashboard isn’t just a real-time view, historical data is stored and accessible, making before-and-after analysis of process changes straightforward
• Eliminates manual observation entirely: no more days of clipboard work, the system runs continuously, capturing data that would be impossible to collect manually at the same scale and consistency

Result

Figure 2: Dashboard for live workshop monitoring.

Makoto can now monitor any client’s production line continuously, without manual effort, from a dashboard that shows exactly what’s happening in real time and what has happened over time. Bottlenecks surface as they form rather than being discovered days later during a manual review, and when process changes are made, the impact is measurable immediately, with historical data enabling an automatic before-and-after comparison. For Makoto’s clients, the quality of the analysis they receive has improved significantly, with recommendations backed by objective, continuous data rather than a snapshot collected over a few days of observation. Manual data collection, which previously took multiple days per engagement, has been eliminated entirely.

Under the Hood