Why a stent engineer rebuilt his cat's dinner

By Prasanna M, founder, Growlrr Foods Pvt Ltd · ORCID 0009-0003-6812-347X
Written in the first person. This one is not marketing copy.

I designed cardiovascular stents for a living — devices that go into a person and must not fail, under loading conditions nobody can fully specify in advance. Then one of my cats stopped being able to pass stool, and I discovered that I had been thinking about his dinner the way an amateur thinks about a bridge: as a recipe, rather than as a system with tolerances.

The five days

In our first year with him, one of my two cats developed a faecal impaction. He spent five days on intravenous fluids. If you have sat in that waiting room you know the specific quality of the guilt — not the sharp kind that comes from doing something wrong, but the dull kind that comes from realising you do not know enough to say whether you did or not.

I had been feeding him carefully. I had read the things one reads. I had a recipe I trusted, and I followed it. And none of that had told me what I actually needed to know, which was: what happens to this diet when the inputs move?

The professional deformation, and why it turned out to be useful

Wondering how this applies to your own pet? Build their bowl — two minutes, and it gives the exact grams for their weight.

In stent design you do not ask whether a device works. You ask over what range of vessel diameters, wall stiffnesses, deployment pressures and fatigue cycles it continues to work — and where the edge of that envelope is. A design that performs beautifully at the nominal case and fails at the fifth percentile is not a good design that got unlucky. It is a bad design that was tested badly.

Standing in that clinic, it occurred to me that I had never once asked that question about a bowl of food. Nobody does. The entire category — homemade, commercial, raw, fresh — is discussed as recipes. This much chicken, this much rice, this supplement. A single point estimate, presented as if the inputs were fixed.

They are not fixed. The chicken from your butcher today is not the chicken from last week. Fat content moves. Cooking loss moves. What your cat actually eats versus what you served moves. Your own measuring moves — nobody weighs to the gram every day, and pretending otherwise is how a plan survives contact with paper and dies on contact with a kitchen.

So I stopped treating it as a recipe problem and started treating it as a tolerance problem. Not does this bowl meet the requirement, but across the realistic distribution of everything that varies, what fraction of days does it meet the requirement, and how confident am I in that fraction? That is a reliability question, and reliability questions have well-established machinery. It is the same machinery, honestly, that I had been using on devices that go inside people.

Ancestral is not the same as appropriate

The first thing that machinery killed was the argument I had found most emotionally persuasive.

The raw and ancestral case is seductive because it sounds like humility — feed the animal what it evolved to eat, stop being clever. I liked it. It is also, on the numbers, wrong in a specific and interesting way.

A wild animal clears its micronutrient floors by sheer volume of food. A hunting cat is burning a great deal of energy and eating accordingly — small prey, many times a day, whole. The micronutrients arrive as a side effect of the calories, because the calories are enormous.

Now take that same composition and feed it to a neutered indoor cat in a Chennai apartment who walks from the sofa to the window and back. He needs perhaps a third of the energy. So he eats a third of the volume — and the micronutrients fall with the volume. Same food. Same ancestral appropriateness. A diet that clears its floors comfortably in a working animal and quietly misses them in a sedentary one.

That is the trap, and it is invisible if you think in recipes, because the recipe did not change. Only the throughput did. Ancestral is not the same as appropriate. What an animal evolved eating and what a particular animal in a particular life needs are two different questions, and the second one is the one in front of you.

This is also why I am wary of any feeding advice that does not ask what your animal weighs and how much it moves. Those two numbers change the arithmetic more than the ingredient list does — and a recommendation that ignores them is not being simple, it is being incomplete.

What I actually built

Once it is a tolerance problem, the shape of the answer follows. You need to know the distribution of every input, not its average. You need to simulate the bowl across that distribution — hundreds of thousands of times, not once. You need to report the result as a confidence statement rather than a tick, because meets the requirement without a probability attached is not a claim, it is a hope.

So that is what exists now: a base of fresh, boneless, cooked meat that you buy and cook yourself, plus one weight-dosed correction for the residual gaps — the things muscle meat cannot supply at any volume, because they are in the skeleton and the organs and the sea. Every version is run against a million simulated bowls, and what we publish is the fraction of days that clear the floors and how confident we are in it. The papers are open access, with DOIs, under a licence that lets anyone check the working. Not because openness is a virtue signal — because a claim nobody can check is not a claim.

The thing I did not expect

I thought the hard part would be the nutrition. It was not. The hard part was the arithmetic of variance, which is a thing engineers are trained on and nutritionists mostly are not, and which turns out to be where the actual failures live.

And I should say the honest limitation out loud: I am not a veterinarian. I have a master's in computational mechanics, I spent years in cardiovascular device R&D, and four years in wildlife conservation before that. What I brought to this was not clinical expertise. It was a specific discipline about failure under uncertainty, and a refusal to accept a point estimate where a distribution was required. Whether that is enough is a fair question, which is exactly why every number is published with its working attached rather than asserted from a position of authority I do not have.

Why any of this should matter to you

It should not, particularly — not in this form. You have a cat or a dog and a kitchen and a limited amount of attention, and you should not have to care about confidence intervals to feed them properly.

That is the whole point. We do the rigor so the user feels the ease.

Build your pet's bowl →