A.I cooked in this kitch -How we obeyed the laws of thermodynamics.

Yes, that’s a money can’t buy Claude shirt. Thanks, Rye (Claude Ambassador). This uh, probably isn’t how I was meant to use it.

So yesterday I used AI to cook, fam. I truly mean it, actually cook. My wife Kasz of justkasz.com fame has a boutique food business, and over the many years she has won many awards for it. But how does A.I come into this? And…

From gold medals, a retail store, and servicing 50 wholesale clients in Adelaide, she packed up and moved to Melbourne. And now stocks a variety of products in some incredibly high-end restaurants. I think she’s up to about seven products going into Vou de Monde (and it’s sister bar Lui Bar) alone with its equivalent of three Michelin stars (the Australian version, the hats). And Ima Asayoru, and Ima’s Pantry, and Georgie’s Harvest, and…okay. That’s enough. We’ll be here a while.

Anyway. Her products are being used at an extremely high level by people that should know extremely high quality, and they do. A couple of products in particular are going absolutely nuts. While she’s always focused on small batch artisanal things, often made in collaboration with the best of the best, a couple of products have gone nuclear that are capable of being stocked on the shelves, and capable of being made at higher volume, at extremely high quality.

Selling out several times in the last few months, including at stockists and online, we need to ramp up production. That means more R&D, bigger R&D, more often. Which gets expensive. So how could we cut back on failures?

Cooking is an act of love, and can be art. But it can also be extremely science based, particularly in manufacturing…and sometimes science isn’t intuitive. E equals MC squared, right, energy equals mass times velocity squared. So when something’s moving twice as fast, it’s actually got four times the energy. It’s moving four times as fast, it’s 16 times the energy. You ofent don’t simply go from “one” to “two” by doubling, in the world of physics.

And we ran smack bang into a question of physics. Quantified with A.I.

In this instance, we were doing R&D on frying things. And while we might use commercial facilities to make product, production research at scale, at home, simply wasn’t going to work.

There’s only so much energy that a home cooktop can put out. Which seems intuitively true. But what this means, for instance, if you’re trying to deep fry something, is there is a hard limit by the laws of physics to how much you can actually do at once.

You might think that you can simply get a large amount of oil super hot, and chuck in all your ingredients while it’s super hot, and it will pretty hot. I wouldn’t jump in that vat, right? So surely this big bucket of oil can do “the thing”, right? Well. No. Not always.

It takes an extraordinary amount of energy to turn water into steam. You’re trying to get all that moisture out of ingredients. This takes energy, and much like when you sweat and heat energy is carried away, cooling you down, the same applies here.

Heating liquid water by 1°C:4.18 kJ/kg
Evaporating water at ~100°C:2,257 kJ/kg

Thats *540 times* as much energy used to boil a kg of water off, than raise the same kg of water by 1 degree celcius. Sheesh.

Between “colder” ingredients going in, and energy turning water to steam, the temperature drops rapidly, significantly. We’re talking from like 105-110 down to like 80-90 in the space of a few seconds. And it can take a long time to come back up. Why?

So, we used A.I to make a graphic to understand what that looks like.

It’s all in the above image you’ve got hot oil. Ingredients go in. You crash below boiling point. Temperature climbs rapidly, but then you see on the blue line, stays flat for a long time, just over 100 degrees. Why? If the burner’s a constant, why are the lines so wobbly?

The blue line is a domestic cooktop burner. Energy is going into the pot, but it’s being used to boil off water, instead of raise the temperature of the oil. There’s only so much energy available, so this takes a long time. All while things are browning, and eventually, burning.

The yellow lines a commercial cooktop burner. Far more energy. Far more violent reaction. Ever seen chips dropped into a deep fryer? You can also see that as soon as the moistures gone, we *rapidly* climb back up to crispy fry temp territory. In fact, there’s a real danger of going too far, and burning things, too fast.

The problem is that around about 100 degrees is not a crispy territory. This is where you’re going to end up with product that’s more poached or forms a crust on the outside and steams itself in the middle. In the end up with chewy, rubbery, sticky things, especially high moisture things like onion and garlic.

So, what’s the solution? Well, on one hand you simply make a smaller batch. On the other hand, you put more energy into the system. That means a bigger burner or a really beefy induction setup. This is where A.I can help tweak and refine, give you rough guidelines on R&D process to try.

In our instance, I used it to do some rough math on timings and temperatures.

• What’s the predicted temperature drop when adding “x” kg’s of “y” ingredients to a pot of “z” litres of “specific oil”, from a thermal mass loading point of view.
• What’s the predicted moisture content of x, and y ingredients, in given quantities?
• How much energy is required to boil that much water?
• What’s the predicted timeframe to boil that with the 4x methods?
• What aromatic compounds and other compounds start suffering, at which temperature points, over how much time? Denaturing proteins, etc?
• What’s the maximum batch size to keep the boil-off period within “x” minutes to prevent over browning, burning, aromatic boil-off?

In this way, we can predict a few things.

• Maximum batch sizes per available facility.
• When a batch of any description isn’t practical on which equipment.
• Predicted time per batch, how many times can we tweak flavours/recipes, in the work time/space available, in a day?
• How many days, time, rounds, KG’s of product, and therefore cost might we need to test “x” ideas, in “y” facility?

And from there, we can understand some costs. The answers helped inform process, as guidelines, no hard rules. They turned out to be extremely close to real world results. I’m talking like within 10%.

So, that’s how A.I helped us with research on a day of R&D, guiding process and recipes before we stepped foot into the building.

Ultimately the one original driver of any product is fun, micro batch creations, tasty dinner adventures, quality, creativity.

Then we get a little help to drive scale.

Just a bit different from the usual SAAS adventures, but helping save time, money, and effort on practicalities. Not guide taste or creativity, literally. A.I as it should be, a tool, not a replacement.

By the way, see something on the site you like? Drop me a line 😉