Backend dev to harden a Claude-built ecommerce app for real clients

I'm building a product that installs an AI growth system into ecommerce brands: ads management, email and SMS, profit and attribution, and a Slack agent that can act on the business. It's a working Next.js app, and I'm about to start installing it into real client businesses. THE JOB IN ONE SENTENCE Everything currently works on my own store. I need someone who can look at it and say "right, this will break the moment someone else uses it", and then fix it. HOW WE SPLIT THE WORK I own the product. I decide what goes in it, how it works and what the experience is, and I build the front end. You make it hold up underneath. Most of your time is backend, the integrations and the data, though you'll touch the front end where something needs wiring up. ABOUT THE CODEBASE (please read this bit) I built this with Claude Code, and I've made a lot of changes to it along the way. It's a real production app rather than a prototype: strict TypeScript, no "any", enums instead of string unions, tests on anything with logic, and conventions documented in the repo. But it has been built fast, by one person who is not a career engineer, with a lot of iteration. Part of what I'm hiring for is a proper pair of eyes on that. If you have strong opinions about AI-written code, this isn't the job for you. If you're comfortable working alongside it and you know how to make it solid, we'll get on well. WHAT YOU'D ACTUALLY BE DOING The code works. It works on one store, with one currency, one timezone, one set of connected accounts, and me as the only user who knows where the sharp edges are. Client installs break all of those assumptions at once. I want someone who reads a feature and immediately sees: - What happens when the access token expires halfway through a sync - What happens on a store with far more orders than mine, or a different currency, or a timezone that isn't London - What happens when a webhook arrives twice, or out of order, or three days late - Whether a refund, a cancellation or a partly fulfilled order quietly corrupts the profit figure - Whether a client could ever see another client's data - What happens when a platform rate limits us mid-run, and whether the job heals itself or just goes quiet Plus the ongoing work: Meta, Snapchat and Pinterest integrations, Shopify Web Pixel and Conversions API, the attribution engine, Klaviyo sync, and guardrails that pause ad spend automatically and must never act outside their limits. THE STACK TypeScript throughout, Next.js on Vercel, tRPC, Supabase (Postgres with RLS), Turborepo with pnpm, Vitest. Meta, Shopify, Klaviyo and Slack APIs. WHAT I'M LOOKING FOR You've worked on ecommerce, or you've built marketing or analytics apps before. I'm not asking you to be a media buyer, and you don't need to know the intricacies of how I run ads. I just don't want to explain the landscape from scratch. You should already have a feel for how the pieces fit together: a store, its orders and products, ad platforms, pixels and tracking, email tools. Anything specific to how I work, I'll explain, and that part is easy. The reason the instinct matters more than the knowledge is that mistakes here are silent. Nothing crashes when a number is wrong. It's just wrong, and a client spends a month making decisions on it before anyone notices. Beyond that: - Genuinely strong written English, because most of our work will be written - You communicate without being chased. If something is blocked, you say so that day - You work quickly and ship. I'd rather have something working that we improve together than something perfect in three weeks - You can own a piece of work end to end, including the tests and the deploy - You'll tell me when the data can't support what I've designed, rather than quietly bending one to fit the other NICE TO HAVE Shopify app or pixel experience, Meta Marketing API, Klaviyo API, or having worked on an analytics or attribution product before. HOURS 10 to 20 hours a week to start, with a view to more if it goes well. Async by default on Slack, with a call when talking is genuinely faster. Some overlap with UK hours. TO APPLY Skip the template. I'll read every application that does these three things: 1. Tell me about something you built that broke in production, what broke, and what you changed so it couldn't happen again. 2. Here's a feature: every 15 minutes we pull a store's new orders and recalculate its profit. Give me three ways that breaks with a real customer. 3. Start your reply with the word "blended" so I know you've read this. I'll reply to everyone who does.