Building a Claude-powered product-listing pipeline behind Viral Printworks
A 12-person apparel company replaced slow, manual catalog work with an in-house pipeline where Claude classifies, structures, and lists print-on-demand products at scale — and engineers itself with Claude Code.
The challenge
Gooder Labs runs a high-velocity print-on-demand apparel operation. Every design has to become a fully built product — the right garments, colors and sizes, on-model mockups, accurate category placement, and clean, search-ready listing copy — and it has to happen fast enough to keep up with weekly drops. Done by hand, that work is a bottleneck: slow, inconsistent, and impossible to scale for a small team. The company had also sprawled into five separate storefronts, multiplying maintenance and fragmenting the catalog.
What we built with Claude
We built an internal listing engine that turns a single design into a complete, correctly categorized storefront product with minimal human touch. Claude does the judgment-heavy work at each step:
- Automated classification and tagging. Each design is analyzed and tagged into a live taxonomy of 23 genres and 35 subgenres, so it lands in the right collections automatically. The taxonomy is synced to the live store, so classifications can’t drift from the catalog.
- Design-first product generation. One product per design, expanded across style, color, and size, with the artwork close-up set as the featured image and on-model mockups and lifestyle imagery assembled in a fixed, quality-gated order.
- Structured, reliable output. Listings are produced against a defined schema with an idempotent design number as the key — re-processing a design updates the same record instead of creating duplicates.
- Catalog architecture. 472 products auto-classified into roughly 60 collections with two-level navigation, tile browsing, and empty-collection guards.
The Claude stack
- Claude API — powers the classification, tagging, and listing-generation logic inside the pipeline.
- Claude Code — our primary engineering tool; it built and maintains the pipeline, and executed a full five-store consolidation across our infrastructure in a single day.
- MCP — connects Claude to our commerce systems and internal tooling so it can act on real catalog data, not just describe it.
How we work
Two engineering habits made this reliable rather than flashy. First, we operate Claude against clear, versioned instructions and documented conventions, so agent behavior is predictable across the codebase. Second, we design for graceful failure — bounded operations, validation on structured output, and re-prompting on the specific error rather than blind retries — so the pipeline degrades safely instead of producing junk at scale. That discipline is what let a small team hand real production work to Claude with confidence.
Results
- A catalog that builds itself. 472 products classified, tagged, and placed into ~60 collections without manual sorting.
- One store instead of five. Five storefronts consolidated into a single Viral Printworks catalog — one checkout, one catalog, one set of collections — with the former brands preserved as collections. The consolidation was executed across our infrastructure in one day using Claude Code.
- Consistency at speed. A modeled library of 95 print substrates and a schema-driven, idempotent process mean every product is built the same way, every time, fast enough to support weekly drops.
- A small team operating like a larger one. Work that would normally require dedicated catalog and merchandising staff is handled by the pipeline, freeing the team to focus on designs and growth.
The live catalog is publicly viewable at viralprintworks.com.