machine vision can help verify delivery orders before they leave the kitchen. Bronze Tracker compares an image of the prepared order with the expected items and provides a check for the team. In chapter 31 of “Toque de Ingenio,” Adrián Martín explains how he developed this proposal, why he simplified the hardware, and what he learned when adapting the product to different restaurant operations.
Guest: Adrián Martín, who was introduced in the episode as the CEO of Bronze.
Interview published: December 10, 2024. Episode: 31. Duration: 1 h 10 min.
In this episode:
- Observing the packaging of orders allowed us to identify a specific operational problem.
- Why a tablet and a simple accessory could actually make deployment easier than a complex camera installation.
- What a picture can verify, and what limitations need to be clearly established for the client.
Adrián’s journey began with a project applying machine vision to a toaster. This then led him to professional kitchens, where he encountered different needs and shifted his focus. The interview details the solution and Bronze’s experiences, which are scheduled for December 2024.
Discovering the problem in the kitchen.
Adrián explains that he needed to enter the kitchens in order to understand how orders were prepared. It’s easy to identify a complaint about a missing drink or garnish from the outside; observing the operation allows one to understand why these errors can occur.
The packaging consolidates recurring tasks throughout the day and can align with periods of high demand. The “Bronze” proposal aims to assist those performing this work by providing an additional check. The product is specifically designed for a particular task, close to the delivery moment.
The conversation illustrates how the concept of business has evolved through a deeper understanding of the client. While the initial technical knowledge doesn’t disappear – it simply finds a new application where it can add value – this evolution requires listening and accepting that the first product envisioned may not be the optimal way to enter the market.
For a team of product development, The approach involves observing the process and formulating a specific problem. A genuine need allows you to determine what the system should detect, when it should do so, and what information the recipient requires.
How Bronze Tracker works.
Adrián presents Bronze Tracker as an application that uses a photograph of the prepared order. The system links this image to the expected composition and provides an indication of visible elements. The aim is to detect any errors or discrepancies before the package is shipped.
The interview describes versions for mobile and tablet use. The first version allows taking a photo in the optimal position for the client’s procedure. The second version can be integrated as a workstation where staff can observe feedback while preparing the order.
The system needs to align with the moment when the elements can still be verified. If the image doesn’t display what’s needed, the analytical capability is limited. The quality of the input data depends on the position, the process, and how the order is presented.
From the design of interfaces and user experience, That relationship is essential. The verification process needs to be integrated into the workflow in a clear and understandable sequence. The person needs to know when to capture the image, what the response means, and what to check for if a difference appears.
A simple accessory to reduce installation complexity.
The development process involved reviewing through cameras and configurations that caused synchronization and connectivity issues. Adrián explained that installing a new client could require travel and technical preparation. This dependence complicated the product’s expansion.
The solution demonstrated in the episode uses an accessory with a mirror behind the tablet to provide a view of the preparation area. The team refers to this as a “periscope.” This approach allows the use of the device’s camera while maintaining a screen accessible to the worker.
The significance of this decision lies in its impact on deployment. A relatively simple physical component helped to reduce the necessary infrastructure. The value of the design is measured by the problem it solves within the system, as well as by its technical complexity.
Adrián describes how simplification changed the way the company approached a new client. The company could now focus more on implementing the changes and adapting them to their operations. prototypes They enabled us to explore various options until we found a combination of cost, usage, and installation that was more practical/convenient.
What the camera can see, and what it cannot verify.
Edgar points to a clear limitation: an exterior image doesn’t allow us to see all the ingredients within a food item or packaging. Adrián acknowledges this and defines the scope of the verification:
“We see what the human eye can see”.
Adrián Martín 16:58.
In this episode, it’s explained that they can distinguish categories when there are sufficient visual cues. Drinks, formats, and labels can aid in identification. If several products are not distinguishable at a glance, the verification process should reflect this limitation.
The application is specifically designed to detect missing elements. Focusing on this aspect allows addressing a common problem without presenting machine vision as a universal inspection of the content. The system requires a definition of categories that aligns with the client’s operational needs.
To develop an AI tool, the case demonstrates the importance of clearly defining the scope. Understanding what can be observed allows for better organization of data capture and labeling. This also prevents attributing a level of certainty to the results regarding features that are not actually present in the image.
A common product for kitchens that operate in different ways.
Adrián explains that the kitchens operated with different processes, even though their ultimate goal was the same. They altered the sequences, the preparation points, and the way in which the elements were presented. The solution needed to find a common core, without requiring everyone to work in exactly the same way.
The pilots and collaboration with kitchen teams enabled iterative development. Observation continued beyond the initial versions, as each new client could reveal further usage conditions. Scalability required understanding what variations the product could accommodate.
The challenge presented is:
“To build something that works for hundreds”.
Adrián Martín 38:05.
The phrase distinguishes a tailored development of a platform that should be reusable. Addressing a single client’s case can provide valuable insights, but the product needs to identify patterns. The architecture should allow for configuration differences without requiring a complete rebuild of the entire solution each time.
To check the order and to retain relevant information.
The interview also covers subsequent complaints. An image associated with the order can provide information about how it left the kitchen. Adrián explains how this documentation is incorporated into review processes with delivery platforms.
That use is different from preventing a lapse before delivery. The former helps to resolve the preparation process; the latter preserves information for analyzing an incident. Both can share a capture, but they respond to different moments and needs.
Adrián insists that the business proposal focuses on improving the restaurant’s operations. Handling complaints is part of this, but it doesn’t represent the full scope of the product’s purpose. The relationship with the customer encompasses speed, quality, and the overall service experience.
The image doesn’t explain everything that happens after leaving the kitchen. Its value is linked to the specific moment it documents. Maintaining that context helps to use the information appropriately and to define what other steps require a review.
Arriving at the market with an understanding of the sector.
Bronze outlines a commercial approach based on relationships with restaurant chains, partnerships, and professional events. Visibility allows for initiating conversations, but understanding the operation is necessary to explain how the system fits in.
The experience of partners and collaborators within the sector helps to identify the most suitable profiles. The technical product needs a way to present itself in a manner that aligns with the business’s priorities. The conversation therefore returns to the starting point: understanding an activity before attempting to transform it.
What can another development team learn?
- Observe the work where the error appears, and identify the most appropriate time to intervene.
- Simplify installation and maintenance, as well as improve the algorithm.
- To communicate what can be verified within the image, and what lies beyond its scope.
- To identify common patterns among clients in order to develop a configurable and repeatable product.
Bronze Tracker demonstrates how a simple hardware decision can simplify a machine vision application. If your company wants to integrate data capture, interaction, and a physical device, i-mas can help you define a solution that fits your operations. Please tell us about the process you need to improve..
To continue reading: Automating an unstaffed shop with Big Fish.
Source of article: Interview with Edgar Guerrero, hosted by Adrián Martín on Toque de Ingenio, published on December 10, 2024. The quotes link to the relevant sections of the video; the “learnings” are a summarized editorial version of the conversation.
