All workLAB-009 / Live product / 2026
Wheerli
See it. Find it. Wear it.
A visual fashion-search system combining garment detection, fashion embeddings, retrieval and multimodal review.
Wheerli · Launch artworkSector
Visual search
Services
Applied AI / Retrieval evaluation / Search architecture / Mobile engineering
Technology
Python / FashionSigLIP / Cloudflare Workers / Vectorize / Cloud Run / SwiftUI / Flutter
Stage
Live iOS product with an active search backend. Android release work continues. Multimodal review is limited to an owner-account rollout; Shopify candidates remain unscored.
Turn a photo into a useful product search, while making retrieval quality, catalogue freshness and operating cost measurable.
Detect and crop garments, retrieve candidates with FashionSigLIP embeddings, then use multimodal review to assess visual matches. Compare stages against a repeatable benchmark before widening the rollout.
Native iOS and Flutter Android interfaces support photo search, crop correction, history and saved results. Catalogue ingestion keeps product records current; review coverage and availability checks vary by result source.
Cloudflare Workers coordinate the search pipeline, with a Python embedding service on Cloud Run and vector retrieval in Vectorize. Application-level budget checks bound search work, alongside scheduled catalogue synchronisation for the live backend.
What we explored
- Retrieval and reranking quality
- Multimodal match evaluation
- Catalogue freshness and search costs
What we built
- iOS and Android interfaces
- Live search backend
- Python embedding service
- Evaluation benchmark and catalogue ingestion
Current stage
Live iOS product with an active search backend. Android release work continues. Multimodal review is limited to an owner-account rollout; Shopify candidates remain unscored.
Next experiment
Add independent human review to the provisional benchmark labels, validate catalogue coverage and continue Android release preparation.
Visit Wheerli