# Digital Twins for Everyday Life

> User-controlled models that remember relevant context, reduce repeated administration, and help the next decision improve on the last

- HTML version: https://robbiepalmer.me/initiatives/digital-twins-for-everyday-life
- Projects: 5

# Vision

> Build trustworthy digital twins of everyday life: user-controlled models that remember relevant context, reduce repeated administration, and help the next decision improve on the last.

Industry builds digital twins for machines, factories, supply chains, and other systems whose state merits a
model. A person's life merits one too. Purchases sit across emails and receipts. Finances span people, accounts,
currencies, ownership, and future obligations. Food decisions depend on who is eating, what is in the kitchen,
what worked before, and how much time or energy is available today.

Most software sees one fragment at a time. The person has to remember the rest, enter it again, reconcile
conflicting records, and translate generic output into a decision that fits their life. They become the database
and the integration layer.

The aim is to give people an increasingly capable model of their own lives. It can start with one useful problem
and widen as new records, relationships, and domains make it more helpful. The limit is not an arbitrary product
boundary or a fear of modelling too much. The model must remain legible, correctable, and under the control of the
person it represents.

Trust does not require keeping the twin small. It requires keeping power with the person represented. A broad
twin that its owner can inspect and direct may empower them far more than a dozen narrow products that each hold
an opaque fragment of their life.

# The shared loop

```mermaid
flowchart LR
Reality["Ordinary activity<br/>records, choices and outcomes"] --> Capture["Capture with little extra work"]
Capture --> Model["Update and expand the personal model"]
Model --> Decision["Support a decision in context"]
Decision --> Outcome["Observe what happened next"]
Outcome --> Model
```

The loop matters more than any particular interface or model. A twin that only collects data becomes another
dashboard to maintain. A recommendation system that does not remember outcomes keeps guessing from the same
thin evidence. The useful system closes the loop and makes routine activity improve the next decision.

This is also what makes the data compound. A purchase can improve later product comparisons. Actual balances can
correct a financial forecast. A completed meal, repeat cook, or abandoned plan can improve the next recipe
recommendation. The person should not have to create a separate diary for the system to learn.

As the twin grows, connections between domains create new kinds of help. A person could understand how food
choices affect their health and finances, how work changes their time and travel, or what a house move would mean
for commuting, spending, routines, and relationships. It could help them coordinate with a household, carry out
plans, delegate work to agents, and attempt things that would otherwise demand too much time or attention.

# How the projects fit

The three Bestomer projects form the beginnings of a commercial twin. The
[Commercial Knowledge Graph](/projects/commercial-knowledge-graph) recovered a person's commercial history from
emails and receipt photos. [Multi-Modal Product Search](/projects/multi-modal-product-search) connected something
seen in a shop or on a screen to current products and prices with almost no typing. The
[AI Shopping Assistant](/projects/chatbot) joined that personal history and product search behind an interface
that could help with a purchase. Together, they connected memory of past commerce, the choice in front of the
person, and the wider market.

The [Personal Finance App](/projects/personal-finance-app) applies the same idea to financial life. Its twin needs
to represent individuals, partners, households, ownership, assets, liabilities, currencies, transfers, taxes,
and plans. It should reconcile forecasts with new observations and answer questions such as whether a household
is on track. The alternative is a spreadsheet whose owner must keep the model coherent by hand and rebuild its
logic for every new situation.

The [Recipe Site](/projects/recipe-site) is developing a food and kitchen twin. It connects recipes and their
lineage with the people eating, dietary constraints, equipment, pantry and freezer state, meal plans, shopping,
portions, leftovers, and what the household actually cooked. That model can answer what will work tonight rather
than rank food by generic popularity. Repeats, adaptations, and abandoned plans then make later suggestions more
useful.

# Toward a connected twin

These projects can become connected parts of a wider twin as they mature. A meal plan can produce a shopping
need. The commercial model can match that need to products, quantities, prices, and substitutions. A purchase can
update both household spending and kitchen stock. The meal that reaches the table can update the recipe's history,
while the financial model can show its real cost within the household's plans.

```mermaid
flowchart LR
Food["Food and kitchen model"] -->|shopping need| Commerce["Commercial model"]
Commerce -->|products, prices and purchases| Food
Commerce -->|spending and ownership| Finance["Financial model"]
Finance -->|budget and plans| Food
```

Product categories should not dictate the boundaries of the twin. Money, food, shopping, work, health, the home,
travel, and learning meet constantly in real decisions. The long-term opportunity is a joined model that can
reason across those boundaries instead of handing the integration work back to the person.

Domain boundaries still help with architecture, provenance, and permissions. They let the owner understand why
the food model needs a budget or why the financial model received a supermarket total. Shared identity,
relationships, and grants can connect the parts while preserving that clarity. People must be able to inspect,
correct, export, disconnect, and delete the model built about them.

# Principles

## Grow through usefulness

The twin has no fixed ceiling. Add a source, relationship, or part of life when it helps the person understand
something, make a decision, coordinate with others, or carry out an intention. Coverage without use creates
administration. Useful connections increase what the person can do.

## Ordinary use should maintain the model

Planning, buying, recording a balance, cooking, adapting, and repeating already produce evidence. The twin should
learn from those actions instead of asking the person to keep a second administrative record.

## Preserve messy reality

People share homes and accounts unevenly. Products change names. Recipes fork. Data arrives late, currencies
move, and plans fail. The model should represent ownership, time, lineage, provenance, uncertainty, and correction
instead of flattening exceptions into one convenient score.

## Work for the person represented

The twin should help its owner make a decision. It should not exploit intimate context to sell the product with
the highest commission, maximize time in a feed, or make leaving artificially difficult. Portability and clear
incentives are part of whether the model can be trusted.

## Connect wherever life connects

The food, commercial, and financial products must each solve a useful problem while they mature. They should not
stay isolated once a real journey crosses their boundaries. The eventual twin should follow the shape of the
person's life, not the shape of the software market.

# Current direction

The Bestomer work tested the commercial twin through working prototypes. The Recipe Site is the most developed
current test because it already records household state and real outcomes. The Personal Finance App remains a
proposal for replacing the spreadsheet that currently holds its model.

The next proof is not the number of facts collected. It is whether each project removes work from the person's
head and makes a recurring decision better. The wider ambition is for those projects to join into a twin that
understands more of the person's life and gives them a greater capacity to act without taking control away.

## Projects advancing this goal

- [Commercial Knowledge Graph](https://robbiepalmer.me/projects/commercial-knowledge-graph.md): Turned emails and receipt photos into structured commercial history, supplying memory for a person's commercial twin.
- [Multi-Modal Product Search](https://robbiepalmer.me/projects/multi-modal-product-search.md): Connected a photographed product to current market choices, giving the commercial twin a low-friction view of what the person was considering.
- [AI Shopping Assistant](https://robbiepalmer.me/projects/chatbot.md): Joined personal purchase history with product search behind a conversational layer for shopping decisions.
- [Personal Finance App](https://robbiepalmer.me/projects/personal-finance-app.md): Proposes a financial twin of people, households, ownership, accounts, currencies, transfers, and plans that can replace a fragile spreadsheet.
- [Recipe Site](https://robbiepalmer.me/projects/recipe-site.md): Builds a food and kitchen twin from recipes, pantry state, household needs, plans, purchases, and cooking outcomes.

---

Markdown index of this site: https://robbiepalmer.me/llms.txt