LoopLook

Inspiration

Most return experiences begin after the shopper has already decided the purchase failed. LoopLook creates one honest, useful moment before an unopened apparel item enters reverse logistics: resolve a color or styling doubt, offer a catalog exchange, or connect the exact unopened SKU and size to a shopper already waiting for it.

What it does

LoopLook begins with the return reason. Physical-fit concerns, damage, incorrect items, hygiene-sensitive goods, and possible reactions bypass AI and continue through a standard return. Color or styling concerns can enter the rescue path after the shopper confirms the item is unopened, undamaged, and eligible.

With explicit consent, the shopper provides a close-up image for optional color guidance and a separate standing image for virtual try-on. YouCam Facial Color Tones identifies skin, hair, eye, and lip colors. LoopLook applies a deterministic, visible scoring model to a fictional retailer catalog. YouCam AI Clothes VTO v3 renders the ordered and recommended colorways as appearance previews.

The shopper can keep the original, request a color exchange, or continue the return. If the return continues, a seeded local relay match lets a second shopper preview and accept the same unopened SKU and size. The retailer dashboard shows the state transition and labels all value, emissions, and packaging figures as demo scenarios.

How we built it

LoopLook is a responsive Next.js and TypeScript application. Consented uploads go to server-only routes, so the YouCam API key never enters client code. The server normalizes asynchronous YouCam jobs to queued, processing, success, or failed states and adds timeouts, recovery guidance, content-hash caching, and a stable seeded fallback for public judging.

The integration uses YouCam Facial Color Tones Analyzer, YouCam AI Clothes VTO v3, a transparent palette engine combining contrast, hue harmony, and shopper preferences, and a guarded return state machine covering keep, exchange, relay, and standard-return outcomes.

We completed one authenticated Facial Color Tones request and two authenticated Clothes VTO v3 requests using original synthetic inputs and fictional Aster blazer references. The public demo uses stable seeded results so judges can complete the journey without credentials or consuming limited API credits. The repository includes the real result images, timings, input hashes, and integration notes.

Challenges we ran into

The hardest product challenge was deciding where AI should not intervene. Virtual try-on cannot determine physical size or fit, so LoopLook routes those returns directly to the standard path and repeats the limitation beside every preview. The main engineering challenge was turning provider file uploads and asynchronous jobs into a reliable journey while protecting a limited API-unit budget. Server-side polling, normalized errors, caching, and a recorded evidence package made the integration defensible and repeatable.

Accomplishments that we're proud of

• A complete six-step shopper-to-retailer journey with valid terminal states.
• One live color-analysis call and two live VTO calls through LoopLook.
• Explicit consent, server-only credentials, privacy disclosures, and recoverable provider failures.
• Explainable color recommendations that shoppers can ignore or override.
• An idempotent relay flow that never claims to move money, issue refunds, reserve inventory, or book shipping.
• A responsive, accessible, credential-free public demo.

What we learned

AI commerce experiences become more credible when their boundaries are part of the interface. Separating color guidance from physical-fit claims made the product easier to trust. We also learned that a deterministic explanation can be more useful than a black-box stylist because the shopper can see how catalog color metadata, detected tones, and personal preferences affected the ranking.

What's next for LoopLook

The next phase would connect LoopLook to retailer order, catalog, inventory, returns, and waitlist systems; replace the seeded relay with consented, policy-aware matching; and validate the scenario impact model with real operations data. Production would also require retailer-defined eligibility rules, fraud controls, deletion workflows, accessibility testing, and measured user research before making business or sustainability claims.

Responsible-use boundaries

• VTO is an appearance preview, not a size or physical-fit predictor.
• Color analysis is optional styling assistance, not attractiveness scoring or identity inference.
• Relay, exchange, retained-value, emissions, and packaging figures are demo scenarios; no transaction, refund, shipment, or inventory reservation occurs.
• The public demo uses seeded YouCam-shaped results; authenticated results and reproducibility evidence are included in the repository and video.

Built With

• ai-clothes-vto
• cloudflare-d1
• drizzle-orm
• facial-color-tones
• next.js
• react
• tailwind-css
• typescript
• youcam-api