Customers & Case Studies
We work with a variety of organizations — sellers, marketplaces, creative agencies and data partners. Below are expanded case studies showing practical problems we solved, our technical approach, and the measurable outcomes. Each case emphasizes reproducible processes and auditability rather than one-off fixes.
Ashley Furniture — large-scale normalization
Problem: Ashley's catalog contained hundreds of thousands of images without extensions, inconsistent naming, and variable internal encodings. The lack of metadata made catalog ingestion slow and error-prone. Our approach combined a forced-PNG pass for consistent downstream handling with targeted byte-preserving fixes for images where fidelity mattered. We generated a per-folder `images_metadata.json` that contained md5, dimensions, derived site/category/subcategory tokens, and a primary-image flag based on filename heuristics.
Outcome: The catalog import workflow moved from manual correction to automated ingestion. Developers spent fewer hours mapping and more time on product features. The metadata-first export also made it straightforward to feed subsets into ML pipelines for product attribute extraction.
Notino Beauty — curated tagging and OCR
Problem: Notino needed consistent tags across fragrance and makeup catalogs, including brand, scent family, and packaging attributes. Many product labels required text extraction for precise matching. Our solution combined tokenization of folder paths, CV label proposals, and optional OCR on high-confidence label images. We preserved original bytes for all critical images and produced an enriched metadata file for each collection.
Outcome: Search relevance and category filters improved, and the operations team reduced manual tagging by more than half. The OCR outputs were retained as raw transcriptions alongside normalized tags, enabling future downstream refinement without reprocessing originals.
Agency partnerships — dataset packaging
Problem: Agencies running seasonal campaigns needed curated datasets with consistent naming, web-ready derivatives, and explicit audit logs. We partnered to define dataset packages (raw, normalized, curated) and developed a delivery flow that produced a ZIP and a JSONL mapping file for traceability.
Outcome: Agencies reported faster creative cycles and fewer rounds of corrections, since every delivered package included both thumbnails for quick review and the high-fidelity originals for final assets.
Want us to feature your case study? Contact us and we'll work with you to publish a short write-up.