FoodChoice: from "what are we eating tonight?" to a household's collective meal memory

What are we eating tonight?

It's a question that comes up every day in a household: what are we eating tonight? The answer rarely depends on one thing alone. It depends on what's in the fridge and the cupboards, on what everyone feels like eating, on what's already been cooked recently, and on a well-known tendency to fall back on the same dishes when what you actually wanted was more variety.

FoodChoice was born from that question, asked in a four-person household.

Starting by remembering

The very first version of FoodChoice did nothing sophisticated. It recorded the dishes that were actually cooked, and kept their history. A few small tools were tried on top of that base, including picking a dish at random from the ones already known.

The catalog describes what you could eat. The history tells you what you actually ate.

That distinction, more than anything else, shaped everything that came after.

When history becomes useful

Once meals were genuinely logged, the accumulated memory started answering concrete questions: when was this dish last made? Which ones keep coming back too often? Which ones have been forgotten?

The clearest example is seasonal: looking up what was cooked for a previous Christmas makes it just as easy to repeat a meal everyone enjoyed as to deliberately try something new. Without history, that choice isn't even a choice — it's an oversight.

From memory to suggestion

The first tools were simple, sometimes random. They gradually gave way to suggestions built on top of the history, mainly to limit certain repetitions and surface dishes that had been neglected.

FoodChoice isn't meant to decide for the household. It surfaces relevant possibilities, based on what was actually eaten — not on a model that would "learn" everyone's taste.

History isn't enough

A suggestion that looks good on paper can be useless in practice if it ignores what's actually available. That's why a more concrete piece of context was added: the ingredients in stock. A suggestion that accounts for what's already at home is worth far more than an abstract one — a point the screenshot below shows directly, with a dish whose stock is already covered.

Choosing together

In a four-person household, real use showed that the choice wasn't always as collective as it looked: the preferences of whoever speaks up most easily tend to win out over everyone else's.

FoodChoice introduced members and individual preferences, then a voting system — still experimental — to make those preferences more visible. The goal isn't only to let the majority win. It's also to make it visible when one person is regularly in the minority, so later meals can make room for them.

A memory that belongs to each household

Food habits, preferences, constraints and the culture around a meal are specific to each household. What "Christmas" means at the table, what comes back too often, what's missing from the stock — none of it makes sense outside the context of a particular group. That specificity is why self-hosting fits the project: FoodChoice remains, today, a private solution, hosted for the household that uses it.

FoodChoice today

FoodChoice today lets you keep a catalog of dishes, record the real history of what was cooked, derive recency, frequency and consumption statistics from it, track a stock of ingredients, and turn all of that into meal suggestions — up to a weekly menu built from several criteria. The collective dimensions (household, preferences, voting) are still being explored and not all of them are stabilized.

One thing worth stating clearly, drawn from real use: FoodChoice isn't opened before every single meal. When the choice is obvious, no tool is needed. It becomes useful when memory, variety, or the household's preferences need to help the decision along.

Build the memory before automating the decision

The most important part of FoodChoice's suggestion system didn't start with suggestions. It started with recording what had actually happened.

That memory is what makes simple but useful questions possible: what have we eaten recently? What keeps coming back? What have we forgotten? What were we cooking around this time in previous years? What can we make with what's already here? What might work for the group?

Suggestions weren't FoodChoice's starting point. They became a consequence of the memory built up through its use.

What are we eating tonight?