DIGITAL FOOD 2.0 By Ajinomatrix

Sep 29

7 min read

Ajinomatrix pre-announces the architecture behind its next generation of sensory intelligence

For years, digital food innovation has been remarkably good at counting things.

Calories. Protein. Sugar. Salt. Ingredients. Purchases. Ratings. Recipes.

But food is not a spreadsheet.

We prepare it. We experience it. We like or reject it. We consume it in a particular quantity, place and moment. It has a physical and chemical composition. It belongs to cultures. It affects our habits. It can be reformulated. And sometimes an unexpected combination appears at the edge of everything we already know.

The digital world has never had a common language for that entire chain.

Ajinomatrix is working on one.

And we believe it could change the way food is discovered, designed, experienced, logged and taught.

From a collection of tools to a food intelligence stack

Over the past years, Ajinomatrix has developed experimental technologies around recipes, sensory digitization, preference prediction and product development.

They are now beginning to converge.

The emerging architecture is based around four fundamentally different questions:

What was made?

That is the domain of JRF: a structured representation of a recipe, its ingredients and its preparation.

What is it made of?

We are investigating a new interoperable composition layer capable of connecting portions and preparations to nutritional, physicochemical and molecular information — including established chemical representations rather than reinventing chemistry.

What does it feel like?

That is MP6 — Multi-Perception 6: Ajinomatrix’s emerging representation of sensory experience across taste, smell, touch, sight, sound and intuition, including how perception changes through time.

And finally:

What was actually consumed?

For that we are beginning work on FMLog — Food & Meal Log, an experimental portable meal-event format designed to record what, how much, when and under what context something was consumed — while retaining links to its recipe, composition and sensory experience.

Four different objects.

One connected food reality.

Recipe → Composition → Experience → Consumption

But that is only the beginning.

What happens when food data starts connecting?

This is where things become considerably more interesting.

A recipe does not exist alone.

It belongs to a sensory neighborhood.

A sensory experience belongs to patterns of preference.

Preferences vary between people, moments and populations.

Ingredients and processes alter sensory trajectories.

Consumption patterns accumulate through time.

And somewhere between all of those relationships are products that nobody has thought of yet.

Ajinomatrix’s SEG — Sensory Experience Graph is being developed to connect those relationships while preserving something that we consider non-negotiable:

provenance.

What a human actually perceived must remain distinguishable from what an AI inferred.

A prediction is not an observation.

A simulation is not a measurement.

A correlation is not causation.

And a rejected AI suggestion must never quietly become sensory “truth.”

This may sound like an implementation detail.

We think it is one of the prerequisites for meaningful sensory AI.

Then comes the strange part: precognition

We use the word carefully.

Not magic.

Not clairvoyance.

Computational precognition.

If enough structured evidence exists around recipes, composition, sensory experience, preference, populations and consumption, the interesting question stops being:

What happened?

It becomes:

What should we investigate before everybody else does?

This is the territory we call:

ABED-SP’G - the datasparagus

An emerging experimental layer combining behavioral evidence, sensory graphs, clustering, decision intelligence and Black Swan PreCog approaches.

Instead of merely finding the center of a dataset, we become interested in its edges.

What recipe sits outside established clusters?

What unexpected sensory combination generates an unusually strong response?

What product looks strange in one market but maps onto an existing preference territory somewhere else?

What apparently unrelated populations converge around the same sensory signature?

And what happens when MP6’s sixth dimension — intuition — is examined as a signal rather than discarded as noise?

Those are research questions.

But they are very different research questions from:

“Which flavor scored highest?”

PRECOG #1 — Discover the drink before the market does

Imagine thousands of emerging latte and iced-coffee recipes being captured and structured.

Instead of reading them one by one, a discovery engine maps their recipe structures and predicted sensory territories.

Clusters appear.

Then something appears outside the clusters.

Is it nonsense?

A temporary social-media curiosity?

Or the beginning of something?

Now add two evolving knowledge layers.

One models sensory preference and likely appreciation.

The other models population and cultural context — not to reduce nationalities or age groups to stereotypes, but to search for evidence-backed sensory affinities across populations.

Suddenly the question becomes:

Who might appreciate this product — including people we weren’t originally looking for?

A concept emerging in Tokyo might have an unexpected sensory-affinity population in Bangkok, Brussels or Ho Chi Minh City.

The interesting unit is no longer simply nationality.

It becomes the transversal sensory ensemble.

This is the direction of the next generation of CACT, WillULikeIT, MP6 and BSPG experimentation.

And eventually the system should be able to ingest real consumer evidence continuously, rather than pretending its first prediction was eternally correct.

PRECOG #2 — AayuSense: from calorie counting to trajectory

Now turn the telescope around.

Instead of:

Which people might fit this product?

ask:

Where is this person’s food pattern heading?

This is the starting point for the new AayuSense PoC.

We will begin deliberately small: a curated experimental collection of health-oriented preparations, foods and decoctions, influenced in part by the extraordinary historical knowledge surrounding Ayurveda — but not limited to India, and without confusing traditional knowledge with validated biomedical evidence.

Users gradually build an FMLog.

Meals.

Drinks.

Preparations.

Quantities.

Estimated calories and nutrients where available.

JRF recipes where available.

MP6 experiences where available.

Then AayuSense can begin producing an index, rather than making the scientifically indefensible claim that a bowl of soup has just added seventeen minutes to somebody’s lifespan.

Think of it as an:

Aayu Trajectory

A directional index indicating how an evolving food pattern relates to selected evidence-backed dietary objectives.

And there is another dimension we want back.

Eco trajectory.

Because the question:

“Is this good for me?”

increasingly lives beside:

“What does this cost the world around me?”

They are not mathematically interchangeable, and we will not pretend that they are.

But presenting personal-food trajectory and environmental trajectory together, with their respective evidence and uncertainty, creates a far more contemporary food instrument than another calorie counter.

PRECOG #3 — Shokuiku becomes a living bridge

Then there is education.

Suppose a school meal can be represented simultaneously as a recipe, composition, sensory experience and actual consumption event.

We can ask:

Where are we today?

Then:

What would improvement toward European objectives look like?

And finally:

What happens if we go further?

This is where the Japanese concept of Shokuiku — food education — becomes particularly powerful.

Not as decoration.

Not as a Japanese label pasted onto a European nutrition calculator.

As a living coaching relationship around food.

Imagine a Belgian classroom connected through the platform to a Shokuiku coach or demonstrator in Japan.

A class examines today’s meal.

The platform helps structure it.

Children explore its ingredients and sensory dimensions.

The coach brings another cultural perspective: preparation, seasonality, attention, balance, experience, origin.

The children change something.

MP6 helps them understand how the experience changed.

And crucially, FMLog can eventually distinguish:

what the school served

from:

what children actually ate.

That difference matters.

A nutritionally optimized meal rejected by half the room has taught us something.

This opens a fascinating possible Belgium–Japan experimental bridge — including future museum, educational and institutional demonstrations — and a particularly compelling conversation around Shokuiku for organizations interested in strengthening Japanese-European innovation links.

PRECOG #4 — Xerox the experience

And then comes perhaps the most provocative experiment.

Take a familiar food product.

Digitize its sensory target.

Understand its composition.

Now impose a different set of constraints.

Less of something.

More of something else.

Different ingredients.

Different environmental objective.

Different nutritional target.

Then ask the machine:

Can you redesign the formulation while preserving as much of the experience as possible?

This turns TasteTuner into an inverse problem.

We are no longer predicting what a recipe might taste like.

We are asking:

Which recipe could produce the experience we want?

Mathematically, the direction reverses:

JRF → MP6

becomes:

MP6 → candidate JRF

Composition constrains the search.

Predicted liking evaluates it.

AayuSense can examine the dietary direction.

The Eco index can examine another dimension.

And ABED-SP’G can search places a conventional formulation process might never explore.

Take the sensory signature of a chocolate-caramel-peanut bar.

Then tell the system:

Make me a substantially improved formulation.

No excuses.

Not identical chemistry.

Not magical equivalence.

But an explicit optimization objective:

How close can we remain to the desired sensory experience while changing what the product actually is?

That is a very different kind of food AI.

The applications are disappearing

There is an irony in all this.

As the technology grows more complex, the interface has to become simpler.

Our recent beta testing made this painfully clear.

Nobody should need to understand five websites, three file formats and a download folder to answer one question about dinner.

So Ajinomatrix is now working toward a unified:

AJX STUDIOS

The emerging journey is radically simpler:

EXTRACT → RECORD → PREDICT → INTERPRET → ADAPT

Behind those five words may sit JRF, MP6, SEG, WillULikeIT, FEELLLM, CACT, FMLog and eventually composition intelligence.

The user shouldn’t have to become an Ajinomatrix engineer to use them.

The intelligence travels.

The context travels.

The provenance travels.

The files remain portable.

The human stays in the workflow.

Four files. One graph. One question.

At the center of the architecture is an unexpectedly simple model:

JRF — What is the recipe?

Composition — What is actually in it?

MP6 — What is the experience?

FMLog — What did we actually consume?

Then:

SEG — How does it all relate?

And finally:

ABED-SP’G — What might happen next?

That last question is where Ajinomatrix intends to spend a great deal of time.

This is a pre-announcement

Some components described here already exist experimentally.

Some are being integrated.

Some are PoCs beginning now.

Some remain research hypotheses.

The proposed FMLog and composition specifications are not being presented today as established industry standards. AayuSense is not a medical device, and its experimental trajectory concepts are not predictions of individual lifespan or substitutes for medical or nutritional advice. Cross-cultural affinity models will require real data, validation and careful treatment of uncertainty. Sensory predictions remain predictions.

We are saying this explicitly because the distinction matters.

But the architecture is now visible.

And once you can represent:

what we make, what it contains, what we experience, what we consume, who appreciates it, where that preference travels, and how the formulation could change…

food stops looking like a collection of disconnected databases.

It starts looking like a computational system.

And perhaps the most interesting question in food innovation changes with it.

Not:

What does the market want today?

But: