77 Roasts Later: Vibe Roasting My Way to Specialty Coffee

77 roasts into my coffee roasting journey, I can say that I have achieved par with the coffee I was trying to copy-roast.

Here are the details of how this started:

The Home Roaster Business Case

These beans started my journey, and I’ve now finally managed to roast the same green beans to the level that the coffee tastes just like the specialty roaster beans I bought.

77 roasts is the total of a whole set of different beans I roasted, and all of them contributed to my learning here.

The coffee I set out to copy took me 17 attempts.

And from the chart below, you can see it pretty much represents the Dunning-Kruger-Effect.

My first batch, when I wrote my last piece on this, was decent. I was happy with the fruity notes, but it was a little dark.

My roaster, the Kaffeelogic Nano, works as follows;

You have a roast profile that’s loaded onto the machine; it describes a roast curve as to how quickly and for how long the beans should be heated before the heating is stopped and stalled. Aside from choosing a profile, you can set the roast level, which influences how high that heating temperature goes and the quantity of beans in grams so the unit knows how much power it needs to reach the temperature.

Naturally, after finding my beans too dark, I went lighter on the roast level, but that just turned the whole thing sour and unpleasant.

That’s the valley I dipped into.

The valley of terrible coffee

As with anything, there’s a learning curve. With coffee, that learning curve materialises about ten days after roasting in the cup you brew.

During roasting, CO2 builds up in the bean and then needs to “degas” for about 10 days to really taste good. After 10 days, many of my failed attempts would either taste grassy and underdeveloped or be too dark, bland, or like instant coffee.

My workflow therefore would be something like this:

1. roast beans with some profile and roast level

2. wait 10 days

3. brew, taste and take notes

4. repeat

You can get really lucky on the first try and head in the right direction, but getting from good to great can sometimes backfire, leaving you worse off by under- or overshooting.

A few lessons I’ve learned here:

I started out roasting 60g batches as I didn’t want to waste precious beans on my first tries. Turns out that the roaster and the profiles aren’t geared for that, and so bumping back up to 100- 120 g solved that.

Ambient temperature plays a role too. As the southern hemisphere winter started, temperatures got colder, and my roasts got worse. Moving the roaster from the garage to the office helped.

Degassing properly. I use peanut butter jars as I’ll have a lot of coffee batches resting and don’t want to buy any fancy valve coffee containers. Specialty coffee beans will come in a pack with a little valve that lets CO2 escape, but keeps oxygen out. That’s handy as it supports degassing without getting the beans stale from too much oxygen. If you just screw the lid of your jar on tightly, degassing will stop, and your coffee will taste somewhat green and underdeveloped.

I’ve noticed this, as usually the first try after 10 days was awful, but then after, say, 17 days it got better. Now, I just lift the lid up once in a while to let the CO2 out.

Finding the right profiles is another challenge. Kaffeelogic comes with profiles, and the community has a whole lot more. With some beans, these default profiles do a good job; with others, they won’t.

Here I found that the company behind Kaffeelogic has a second, more expensive roaster targeted at professionals. Apparently it’s almost exactly the same, just with some different software and different profiles.

Turns out they did a sloppy job of encrypting these, so I managed to get all of the Nucleus Link profiles and made them compatible with Kaffeelogic, and the results are quite good.

The Nucleus Link profiles aren’t a different roaster so much as a different roast strategy on the same fluid-bed hardware. They just seem to be geared towards the specialty coffee I’m trying to roast.

Vibe Roasting

Now obviously, this would have all taken me years, or at least consumed a lot more of my time, about five years ago. But as any sane person with a large token budget, I built some tools and frameworks to help me vibe roast my way to shorter feedback loops and better coffee.

The first thing I needed was tracking the different green beans I have and saving their profile so that I can feed that context into a prompt.

Aside from the green bean profile, the prompt receives past roast logs of other coffees, the roast profiles from the Kaffeelogic machine, a guide on how the profiles work and all of my cupping notes from when I tried out a coffee.

So when I plan a new roast, the AI model will have information such as “that last cup was a bit sour and looked a little dark”.

I don’t necessarily know what that means for roasting, if the profile is right or what level it should be, but with all the context, I usually get a decent next step that improves upon the last one.

For every roasted coffee, I’ll track the logs from the roaster, the weight before and after and the time it took. Then I’ll keep track of resting time to see how long it takes.

This makes the workflow so much simpler and lets me iterate on coffee much faster.

I don’t want to throw away beans, and I don’t want to read endless books about the theory of roasting, so collecting data and having an agent assist you in the process makes a ton of sense.

AI, the hobby accelerator

Reaching parity with Rocket Coffee was why I bought this roaster. Achieving that makes me feel like a roasting pro!

Without the structure and the app, and with an AI model in the loop to help me navigate all the variables, this would have taken me so much longer. Honestly, I probably wouldn’t have stuck with it.

The Kaffeelogic Nano is a great system, but home roasting used to demand a serious time commitment. To do it the traditional way, it basically needs to be a proper, full-time hobby.

What’s incredible about where technology is today is how AI lets you speed-dial through learning something that previously felt out of reach. It took a high-friction, time-consuming hobby and turned it into something I could easily pick up and master in my spare time.