Ai roasting and the future of micro roasting

Coffee Talk
1

Recently saw a deleted thread titled “ChatGP can’t roast” and it got me thinking.

We all know that to make a good cup of coffee, it can take many years of experience to become good at. this is because of the many variables that make up the flavor chemicals that are present in a cup of coffee.

A.i. has never been a viable option for roasting in the past, and part of the reason was how expensive data is. You would not be able to employ all of the necessary people in all of the places around the world to use the same machine….

However with tech today, you can actually get people to pay for the roaster, and give you all the roasting data for free. Sort of the tech industry approach to customers, where you make the customer pay for the product, and you use the customer as a data stream to then again repackage and sell. The off shoot challenge is getting much dirtier data, as no one has to fill in all the data parameters to make it easy.

But can Ai make art? Can Ai make good flavors?

my assumption, base on what i am seeing in the music industry and visual art iterations, ChatGP, architecture, and so many other segments, that yes, Ai roasting will happen.

What will the fall out be? Chances are it will remove the need for majority of micro roasters that are on thin margins. So, if you do this for the fun, and profits are small, Ai could squeeze your margins even less. Because you’re not just competing with other hobbyists and professionals, you now will be competing with anyone who can afford a machine roaster. In the short term this will be great, as you’ll be able to expand with fewer employees, but in the long term, you are not as unique.

from what i’ve gathered in working in roasting and chatting with the individuals here, is that any number of roasts can be delicious.

With machine learning data collected at Aillio, they are now amassing thousands and thousands of hours of roast data from around the world. Not that the data is directly applicable to begin with, but once you get into the orders of magnitude greater data, then suddenly you have something that could be worked with. We’ve seem this play out in how ChatGP works, although that is a lot more data.

We are currently in a coffee renaissance, as more people than ever before are able to do it, and many even make a little money on the side, while others are able to turn it into a growing business. The talent pool for roasting, and the inter communication between roasters is at an all time high. Incentives are also currently rewarding for those who pursue it well. We are at a point where we can all learn and grow very quickly. But it takes sharing and talking and it takes data. Will we get to the point we want before Ai makes inroads into profit margins?

So, what are your thoughts on the future of micro roasting for a living? Can a bot roast well enough? Would you prefer to leave the roasting to ai? are you going to roast as a hobby even if there is not enough money to support a family? Or will coffee roasting always be a human operated thing? Too many variables for Ai to handle it well. Where are you optimistic and see opportunity in Ai?

Thanks for reading and interested to hear your thoughts!

1 Like
2

So some context: I studied data science and have employed ML models before.

In my opinion, AI is a tool, not the end of roasting as a career.

In the context of Aillio specifically, we are using ML to help roasters roast better. We are not trying to replace anyone. Can it help you roast better? We hope so. Can it create it’s own roasts? Sure. But there still is going to be a roaster validating profiles, determining what constitutes a “good” roast, and experimenting with different inputs.

Maybe there is some misinformation about our new machine and the data we collect. I want to make it very clear right now, we are using this data to help roasters improve their roasts. Not cut them out. Maybe in the future the roaster could be monetizing their recipes and profiles instead of spending hours doing the roasts themselves.

4 Likes
3

Hello Jimmy,

I don’t think AI could replace a human roaster, the reason is, roasters keeps adjusting their profiles till they get what they want based on tests sone using their abilities to taste and smell, if there was any Ai able to do such things, it would had been used in food industry. AI can be used to ease the work flow, detecting FC as an example.

2 Likes
4

To say what both @mcaillio and @alghamdi.majid1b1s said slightly differently.

ChatGPT works by “reading” a huge amount of text so that it can generate intelligent-seeming replies to prompts based on the massive (~175 billion) parameter network that it built from the text. All it needs is the text for this.

DALL•E works in a similar manner, except that it also “reads” images and captions, so that it can “reply” in the form of synthesized images.

I don’t think that the significant results here translate to coffee roasting, yet. The data for a language model is text, and it serves as the input and the labeled output for training. With coffee, the output isn’t text, or an image – it is a cup of liquid with taste, smell, and mouthfeel. You can gather gigabytes of roast curve data, but without high quality labels that indicate those characteristics, I’m not really sure how you build a model that automatically roasts coffee to some target preference.

The same problem applies to the input as well. Text is text is text, but your green coffee is not identical to the “same bean” that a model was trained on, due to the environmental factors affecting the coffee from the time that it leaves the grower to the time it drops in the roaster.

Then there is the roaster itself - two “same models” of a roaster may have different gas and airflow setups. All of this would need to be accounted in the model parameters.

Not saying it can’t be done - but doing something along the lines of “ChatGPT for coffee” doesn’t have an ROI yet. As mentioned above, that doesn’t mean that machine learning can’t be used as a tool to help roasters more easily achieve consistently good results. I’m pretty sure we’ll see a stream of innovations along these lines as the Cropsters of the world get more and higher quality data.

2 Likes
5

Is there a preference to coffee? I mean is the taste of what’s in the cup objective or subjective?

And i’ll also ask, do you think most roasters are good?

I’ve seen some production roasting, and I have yet to see that there is some inspiration that they apply that a machine cannot.

So instead of having a team of three roasters, one could just have one roaster doing qc, and the machines automating the rest. in particular blends. The results would be more consistent than humans.

where machines might fail in the past is having the ability to adapt. Partly do to lack of data.

6

We’re already seeing machines replace human roasters. What won’t be replaced any time soon is the qc or taster.

I’ve met a few people that like the coffee coming out if the machine better than some of these local roasters. And i don’t think they’re wrong for it.

7

Hello Jimmy,

What in the cup subjective as a consumer, but as a roaster it is an objective. First rule in Roasting business is understanding the market you are targeting. For example, I like the fruity coffee. Thats subjective I like fruity, you may like dark strong coffee. But If I had a chance to work in Italy, Italians are not fans of speciality coffee, they are mostly into Arbica blended with Robusta, I will roast what suits them and that an objective and thats how I will roast as a business.

The second part, do I think all roasters are good!? In the business of roasting, No, and thats normal in any field, when things come to money, money doesn’t lie. When you visit me and I invite to a cup of coffee I roasted, you may say it was good coffee as a complement, even if it wasn’t that good. But as a customer, someone pays money, they will not buy from you unless your roasted coffee was good and they get what they are paying for. And these not good roasters, will find themselves out business.

2 Likes
8

What is your profession? Are you roasting for a company? Are you an educator? Where do you want to end up?

9

Hello Jimmy,

I work for aviation field, I am not an educator yet(I plan to get the SCA AST this year), I cooperate with local roasters in designing their profiles, but I don’t own my own roaster yet. I want to have my own business in coffee and be among the best roasters in world.

10

I think it is safe to say that a well funded and highly talented and driven roaster has nothing to fear of AI in the next 20years.

Where it is going to have a bigger impact is all the middle and low end micro roasteries. The small upstarts. And there are plenty of people tha absolutely love coffee roasting that are not good at it. We’ve all just had to deal with it, because of the need for them. If a robot can do better, we’ll see certain types of business models drop off.

The question in the original post was different than the direction you three guys have taken. I think MC was being a good person and trying to stand up for the work he does. that’s great, but it did end up shifting the conversation towards something extreme. Very all or nothing.

The truth is, that automation and machine learning will have an impact on the number of required roasters. If cafe’s offload the roasting to their lead barista, you only need a few cafes to transition before a micro roastery is deeply effected. as it is already quite competitive and saturated.

The solution is not to stop Ai, but to consider that it is happening. And if you aren’t the best of the best, what do you do?

11

I recently came across a Bellwether automated roaster in a small coffee shop, maybe as close as you can currently get to AI. I don’t want to admit, but the coffee tasted good. The roaster is completely self contained and requires NO venting. The future of roasting?

12

The Bellwether isn’t really AI, it’s just highly automated. The roasts are all developed by experts at the factory and Bellwether sells the stores the green beans. All the operator has to do is download the roast instruction and dump the beans in.

An old friend works for Bellwether and is helping develop the newest version of the machine.

With good beans, the coffee is pretty good - the roast developers are pretty bright. I was on the coffee gravy train for a while - my buddy was doing final commissioning on the older version that was assembled here in Portland. Every machine that came off the line had 5lbs of coffee put through it, so he had LOTS of coffee to give away. I bought my Bullet when they moved manufacturing to Berkley!

I think the AiO is going to give the Bellwether a real run for its money. It might be a bit smaller capacity (2 KG vs 5 lbs), but it’s less than quarter the price and doesn’t require three phase power.

I think induction roasting is a real game changer. It is so much more efficient than heating up air with resistance heaters, so the electrical infrastructure to run it is considerably cheaper and you don’t have to have access to natural gas, nor all the fiddly regulations surround high input natural gas machines.

-Gray

2 Likes
13

Dear Larry,

Kindly check the screenshot taken from their website :

image
image750×1334 118 KB

The profiles either done their master roasters if you pic a pre-made profile or you customize your own profile. And pre-made profiles can be applied on beans purchase through their Marketplace.

So, as @gfh65wi said, it is not an Ai toaster. The fully automatic roasters are available in many shop roasters, automatically loads the batch, charge it to the roasting chamber, follow the designed profile (custom made or integrated) and drop the beans to the cooling tray.

1 Like