Nathan Joel

Every brand arrives with its own catalogue, house style, languages and quirks. I take them from that first scoping call all the way to live releases — pulling their product data in, shaping the rules that fit their voice, checking the output, and shipping it in large batches without any one client turning into a special case.

Most of the day-to-day is keeping a lot of plates spinning at once: standing up a new brand, clearing stalled work, handling the awkward edge cases that only appear at scale, and staying close enough to each account to catch the requirement nobody thought to write down.

Good product copy starts with good data, and retailer data is rarely clean — specs hidden behind dropdowns, the same attribute named five different ways, dropdown options that look like features but aren’t. I build the pipelines that pull all of that in, untangle it, translate it across languages, and hand the AI something it can actually work with.

On the generation side, the system writes titles, descriptions and metadata in stages — research, extraction, drafting, checking — so each step can be tuned without breaking the others. I ship full-stack features across it: the tools operators use to run and review jobs, and the quality checks that catch the failures you only see at volume.

When the same problem shows up on every other account, it’s usually worth solving once, properly. I wrote the design proposal for a flexible data-export layer — per-client mapping handled as configuration rather than one-off code. It was adopted as the product direction and now removes a whole class of bespoke work from every new onboarding.

The same instinct shows up in smaller tools: taking a fiddly one-off client workflow and turning it into something reusable the whole team can run, and fixing the cross-cutting bugs that quietly block everyone.