Identifying Coffee Blends Through Aroma Detecting AI Using Calyx’s Customizable Sensing as a Solution Platform
The Challenge: Acme* wanted to ensure the quality of all the coffee being processed at their distribution facility. Historically, sensing solutions for coffee aroma have been limited, as there are many volatile organic compounds (VOCs) across blends of the bean. Acme had exhausted their existing solutions and failed. They came to Calyx and asked us to create a new customized Sensors as a Solution* platform for their coffee quality assurance and blend personalization. We had a concept for how this could work, but until Acme presented us with the problem, we’d yet to prove our solution was the best.
The Approach: We started by filtering materials in our receptor database to find an array of sensors that, when combined together, produce a unique response for each aroma of interest. Then we built a unique “map,” or profile for each aroma blend. This profile is key to the statistical techniques that we use to compare a new blind sample against these profiles, and is how we enable automatic identification. This is uniquely agile, as our proprietary technology and approach makes it possible, even when the full list of compounds specific to an aroma remains unknown.
Smelling Coffee Aroma with Sensors: So how exactly does this work? Once we identify the aroma profile that we want to detect accurately, we create a custom sensor end-to-end solution (hardware and software.) For Acme, we put together an array of sensors, and depending on the complexity of blend aroma, decrease or increase the number of sensors in the array.
Each sensor is made with a different receptor on it, making them different from another in what they can sense, and/or degree of response towards certain gases. Upon exposure, the interaction with chemicals in the blend cause a physical parameter change on each phage sensor that lead to a change in the visible color. From similar interaction events across all our sensor arrays, we can get a profile of all the different color changes, which we will record in our aroma library. The library can later help us identify how well the particular aroma matches any of the recorded profile in our library, which will in turn tell us what aroma it is, and how similar the sample is to each recorded aroma profile.
Once the aroma sensor is trained, it will take just one attempt to get the sensing right. However, the training (profile recording) is the time-consuming part, as the profile being generated for different aroma must be correct and we need to run multiple times to confirm.
With our aroma sensor, we can iterate very fast, because if the sensor array cannot tell the difference between targeted aromas, we simply swap out or add in more sensor modules into the array. This change can happen as needed, which allows us to continuously add more capability to our analyzer.
Future Implications: Beyond coffee, we could apply a similar approach to sensing tea, chocolate, frozen food, perfume, and so much more. And in theory, we could correctly sense all the different coffees on the market, as our receptor database library has more than a trillion receptor options (significantly more than conventional technologies.) To say we could sense them all today would be a stretch. We are actively building out our library and training our system to increase our ability to sense unique blends of the world. So far, the results are promising. We look forward to the next coffee challenge our community presents us with!
If you’re interested in speaking with our team directly, please reach out to info@calyxtechs.com.
