Cloth-ee

Inspiration

Clothing waste rarely begins at the trash bin. It often begins earlier when someone sees a post, follows a trend, buys a cheaper copy, and replaces clothing that still has years of use left.

Cloth-ee originally started as a virtual fitting room. While building it, we realized that placing a digital shirt over someone did not answer the more meaningful question:

Why do people buy, replace, and discard clothing, and what could change that behavior?

That realization transformed Cloth-ee into a transparent fashion behavior simulation. Instead of claiming that we can predict a real shopper, we created a synthetic town where assumptions are visible, decisions are traceable, and interventions can be tested.

What it does

Cloth-ee simulates a town of 40 residents with different:

Budgets and incomes
Fashion preferences
Jobs and routines
Social connections
Brand loyalties
Sensitivity to trends
Interest in sustainability evidence
Disposal habits

Residents move through town, visit stores, purchase garments, wear them, influence each other, and eventually resell, donate, or discard clothing.

Every decision follows inspectable rules. When a trend causes a purchase, Cloth-ee can trace the complete chain:

Social signal → purchase → displaced garment → unused wears → disposal

Users can select a resident to understand what they are wearing, why they bought it, and whether another garment was replaced prematurely.

Cloth-ee also runs controlled experiments. It creates two copies of the same seeded town and changes one signal, such as whether a brand publishes a clothing product passport.

It then compares:

Total purchases
Purchases from the selected brand
Items sent to the bin
Prematurely discarded garments
Potential wears lost
Resale and donation rates
Trend related textile waste
Differences between shopper personas

The same seed and settings always reproduce the same result. This means the changed signal, rather than unrelated randomness, causes the difference between the two runs.

Cloth-ee presents these outcomes as synthetic scenario results, not predictions about real people.

How we built it

Cloth-ee is a browser based simulation built with JavaScript, HTML, CSS, Canvas, and a Node.js backend.

The shared simulation engine models:

Resident schedules and movement
Shopping decisions
Product prices and availability
Social posts and conversations
Trend growth and decay
Brand awareness
Garment durability
Wardrobe wear counts
Premature replacement
Resale, donation, and disposal
Product passport availability

Each resident gives a garment a purchase score:

$$ \text{Purchase Score} = (\text{Style Fit} \times \text{Style Priority})

(\text{Evidence} \times \text{Evidence Priority})
\text{Need}
\text{Trend Influence}
\text{Social Awareness}
(\text{Price Pressure} \times \text{Price Sensitivity}) $$

Different residents assign different importance to these signals.

A Conscious Buyer responds strongly to product evidence. A Trend Chaser responds more strongly to style and social momentum. A Bargain Hunter is more sensitive to price. A Minimalist has less interest in shopping and keeps garments longer.

The interface renders a living town and displays purchases, posts, conversations, wardrobe changes, and disposal events as they happen.

We also built:

A deterministic baseline versus intervention runner
Product passports with explicit unknown fields
Material based lifecycle calculations with stated boundaries
A Shopify Storefront adapter for products, variants, carts, and checkout
An OMNI multimodal adapter for visible garment inspection and spoken questions
A local visual matcher that can be evaluated against simulation ground truth
Automated tests for determinism, evidence handling, catalog matching, sessions, and API behavior

The language model does not control resident behavior. Resident decisions remain rule based, reproducible, and testable. OMNI is used on demand to inspect rendered pixels and explain information conversationally.

Challenges we ran into

Finding the real product

Our first concept focused heavily on virtual try-on. We experimented with body landmarks, garment overlays, sizing profiles, and augmented reality interfaces.

We learned that reliable virtual try-on requires more than recognizing a person. It requires garment specific assets, body measurements, depth, segmentation, occlusion, cloth deformation, and validated fit models.

Presenting a rough overlay as accurate fitting would have weakened the project. The more interesting system was underneath it: behavioral signals, clothing lifecycles, social influence, and the consequences of replacing garments too early.

That led us to make the simulation the core product.

Making behavior understandable

It was easy to generate purchases and disposal events. It was much harder to explain why they happened.

We added causal records connecting social posts, trends, purchases, displaced garments, remaining wears, and disposal routes. This transformed the town from an animation into an inspectable experiment.

Avoiding misleading sustainability claims

Lifecycle information is incomplete and often uses incompatible system boundaries. A cradle-to-gate carbon estimate cannot be compared directly with a complete lifecycle assessment.

Cloth-ee separates information into three categories:

Events observed inside the simulation
Estimates calculated from sourced factors
Unknown or unsupported information

If a required factor or source is missing, Cloth-ee shows it as unknown instead of asking a model to invent a number.

Comparing scenarios fairly

Two simulations can produce different results simply because of randomness.

To isolate an intervention, both scenarios use the same residents, schedules, social graph, products, and random seed. The intervention is the only changed signal.

Managing scope

The project evolved through virtual fitting, commerce, multimodal AI, product passports, lifecycle analysis, and behavior simulation.

Our biggest challenge was deciding which story these systems genuinely supported.

The final answer was to use the town to make clothing behavior visible, use controlled experiments to test interventions, and keep every claim honest.

Accomplishments that we're proud of

Built a living simulation with 40 synthetic residents
Created deterministic and repeatable scenario comparisons
Traced trends from social posts to purchases and premature disposal
Modeled garments through purchase, wear, resale, donation, and disposal
Made resident decisions inspectable instead of hiding them inside an AI model
Kept missing environmental and ethical information explicitly unknown
Allowed OMNI to inspect rendered pixels without receiving hidden outfit data
Created top-1 and top-3 garment matching evaluation
Connected product passport visibility to evidence sensitive behavior
Built an interactive town that works without external API keys
Added Shopify product, variant, cart, and checkout support
Reached 22 passing automated tests

We are especially proud that Cloth-ee does not force every intervention to look successful. An intervention can improve one metric while making another worse. The application reports that tradeoff instead of manufacturing a positive result.

What we learned

The most important lesson was that simulation and prediction are not the same thing.

Cloth-ee does not prove how real consumers will behave. It helps formalize assumptions, expose possible mechanisms, and identify ideas worth testing with real data.

We also learned that:

Sustainability evidence matters most when it appears during a decision.
A traceable story can communicate more than an isolated carbon score.
A garment discarded with 40 possible wears remaining makes waste understandable.
Deterministic systems are easier to test, explain, and challenge.
Language models are more useful when grounded in observable information.
Missing data is meaningful information.
A smaller honest claim creates a stronger product than an impressive but unreliable one.
Product development sometimes requires abandoning the original interface while preserving the strongest underlying idea.