An AI Operations Platform for a Two-Country Business

The business runs across two markets — Japan and the US — but the product pipeline between them was entirely manual. Japanese product descriptions needed translating, images needed renaming for SEO, pricing needed localizing per variant, and tracking which products were live required constant coordination. None of this was systematized. It all lived in emails, messaging apps, and verbal agreements between me in the US and the owner and ateliers in Japan. I was the bottleneck for every step.

• 01

  I sat down with the owner to map the workflow end to end before building anything. The scope was immediately clear: a large product catalog, each product with multiple SKUs, all described in Japanese, none of it translated, none of it organized for a US market. There was no system tracking what had been done, what was pending, or what was live. Everything existed in someone's head.

• 02

  The existing tools weren't the problem. The owner was already using Google Sheets with ateliers and suppliers, that workflow worked for them. The gap was everything that happened after: taking product data from that sheet and turning it into something ready for a US Shopify store. Translation, image naming, SEO, pricing, status tracking. None of those steps had a home.

• 03

  The core constraint was that I couldn't add new tools to the owner's workflow. Any system I built had to live inside what they already used or run invisibly in the background. That ruled out a custom interface for the owner entirely and pointed directly to the architecture: sync from Sheets, automate the localization layer, surface the results in a dashboard the team on my end uses.

• ·

  The owner and ateliers manage products through Google Sheets, which is already shared across suppliers. Rather than replacing that workflow with a new interface, I built the dashboard to sync from it. The owner updates product info in Sheets and the changes propagate to the dashboard automatically. The complexity lives in the tool the team already knows.

• ·

  The dashboard handles what Sheets can't. When a product is ready, clicking one button triggers Claude to translate the Japanese description into English using a structured template, and generates SEO-optimized filenames and alt text for each SKU variant.

• ·

  I used Nano Banana Pro for product image generation with brand kit constraints. Locking style parameters to the brand kit meant consistent visual output across sessions, which reduced the need for manual review and rework on every new product.

• ·

  The business owner kept switching between tools to check performance. I noticed the pattern and added an analytics layer that pulls from Shopify, Instagram, TikTok, and Pinterest into the dashboard. It became one of the most used parts of the platform.

The platform handles product translation, image management, and status tracking for both markets from a single dashboard. Products move through a clear pipeline from handoff to live on the US store. The analytics layer surfaces Shopify revenue, social performance across three platforms, and top-performing products in one place. Coordination overhead dropped by an estimated 65%.

• ·

  I was both the designer and the only engineer, which made it easy to skip proper discovery. I built the brand kit system based on what I thought was needed, not on a structured conversation with the stakeholder about what actually caused inconsistency in his eyes. It worked, but I got lucky. A proper stakeholder interview upfront would have been faster.

• ·

  The hardest part wasn't the automation. It was making the status pipeline legible to everyone using it. The owner and ateliers needed to understand at a glance where each product was without logging into the dashboard at all — which is why the status columns write back to the Google Sheet they already live in. That decision came late and should have come first.