COFFEE & ARTIFICIAL INTELLIGENCE
BKI creates coffee recipes with artificial intelligence and wins the Danish AI Award 2022
In collaboration with The Danish Technological Institute, BKI has developed a statistical model that can predict coffee taste and formulate coffee recipes with artificial intelligence. An innovation that led us to victory at the Danish AI Awards 2022 and a tool to help us create your coffee blends even better.
On August 25th, 2022, a small team from BKI and The Danish Technological Institute entered the stage at The Danish AI Awards 2022. An event hosted by the Danish Industrial Foundation and AI Denmark during the annual Innovation Festival.
Here, a unanimous panel awarded BKI the victory trophy and a prize of 200,000 DKK for the best AI project of the year among more than 75 participating Danish companies.
But a project that formulates something as subjective as taste is not something you see every day, they said. Especially not when it comes to such a complex and nuanced food like coffee.
Through one and a half years of close collaboration with The Danish Technological Institute, we have managed to develop a statistical model that, with artificial intelligence (AI), can predict the taste of combinations of raw beans and then generate precise recipes for our many coffee variations.
Before they are even roasted.
A tool that will help us become even better at creating coffee blends and ensure that it tastes exactly as our customers know and loves them.
Now, just even smarter and with greater flexibility in the purchasing process.
"This is completely going to change our purchasing process of raw coffee. Now that we can generate coffee blends based on a predefined flavor profile, with handpicked and specific criteria added to it, we are no longer as dependent on seasons, specific coffee varieties, or coffee regions, as we were before. Now, I can purchase and create coffee blends much more flexibly."
- Casper Rasmussen, Purchasing manager and expert taster at BKI.
ABOUT THE AI PROJECT
- AI stands for Artificial Intelligence.
- Our AI model is based on taste assessments of raw beans and can predict the taste of a specific coffee blend and generate precise recipes for it.
- The Danish Technological Institute has been involved in developing the statistical aspects of the AI model, while BKI itself has been responsible for developing and implementing the necessary tools.
- BKI won the Danish AI Award 2022 for the project, along with a prize of 200,000 DKK from the Industrial Foundation. These funds are earmarked for further work with AI at BKI.
In the video, Ib Hauberg, former factory manager in BKI, provide more details about the AI project (in Danish).
From analog to digital coffee tastings
For over 25 years, our purchaser and chief coffee cupper in BKI, Casper Rasmussen, has relied solely on craftmanship, knowledge and his taste buds to ensure the flavor and quality of our coffee blends.
It's a craft he is very proud of, but, as he puts it, it has been very 'analog' until now.
"Where I used to write down all my coffee tastings on paper, I now type my assessment of flavor parameters of raw beans and roasted coffee blends onto a special app that we developed. And after seeing what our AI model can extract from these data, I've only become more eager to continue working with this," says Casper Rasmussen.
And even though BKI's coffee tasters are already extremely skilled at creating our well-known coffee blends with trained taste buds only, artificial intelligence simply opens even more possibilities in the daily coffee production.
Now, Casper Rasmussen can request the AI model to create detailed recipes for a desired flavor in a finished coffee blend.
Before the beans are even roasted.
Casper Rasmussen, who is the head purchaser and expert cupper at BKI, tastes an average of 300 cups of coffee per day. This includes samples of raw beans and roasted coffees. Each time, he assesses the coffee on a wide range of flavor parameters, which he enters directly into an extensive coffee database. This data is the foundation for the AI model.
But it's not just the taste he can ask the model to account for.
He can also define a price and quality for the coffee, a specific content of particular coffee beans, or he can ask the model to create the recipe from the raw beans available at BKI at that particular time.
"This is completely going to change our purchasing process of raw coffee. When we can create coffee blends based on a predefined flavor profile and handpick specific criteria, we are no longer dependent on seasons, specific coffee varieties, or coffee regions. Now, I can purchase and create coffee much more flexibly - and even ensure a more consistent coffee flavor and quality for the consumer," says Casper Rasmussen.
"That is incredible.”
But before it was even possible to benefit from the predictions of the AI model, a lot of work had to be done.
Namely, the work of collecting and connecting the right data.
Because without the right input, there is no output from a so called ‘intelligent model’.
Module 2:
Evolutionary algorithm
The evolutionary algorithm is the part of the model that BKI puts to work. By requesting the model to create a desired flavor profile for a coffee blend - and, for example, linking variables such as price, coffee varieties, and roasting degree - the algorithm will generate a wide range of possible recipes for blends of raw beans. It continuously predicts and optimizes the flavor profile using data from the neural network.
Coffee database is the foundation for artificial intelligence
While experts from The Danish Technological Institute have been our collaborators throughout the AI project, BKI itself has been responsible for designing and implementing the necessary tools to make the use of artificial intelligence possible.
And Nicholaj Månsson Olsen, data architect at BKI, is the man behind the apps that are now essential tools in our coffee production.
"I've spent several months developing a range of apps and a digital setup that both suppliers, our coffee cuppers, the roastery, and packaging department, all can access and record the necessary data for the model,” says Nicholaj Månsson Olsen.
“A digital coffee database.”
And to connect all this data, the need for artificial intelligence arises, and not just for a regular computer.
"In BKI, we typically have between 50-70 different types of raw coffee beans in stock for our coffee blends. Combined with production parameters, such as roasting levels, this means that we have a theoretical possibility to create and roast our coffees in no less than 99 trillion different ways," says Nicholaj Månsson Olsen.
An astronomical number of possibilities that no human brain can comprehend at once - despite of trained taste buds.
However, Casper Rasmusen, the purchaser and expert cupper at BKI, is not worried about artificial intelligence making him redundant anytime soon.
According to him, the AI project has, in fact, made him even better at his craft.
"Our AI project has only improved the quality of our tastings - I have become better. The new procedures force me to sharpen my senses and evaluate the coffee based on far more parameters than we are used to. And I think it's fun. We have revelations all the time and we are constantly coming up with new exciting ways to translate the data into better coffee."