It works on the demo. Not on real users.
Edge cases, concurrent writes, weird inputs — the moment real traffic arrives, things start falling over.
AI code rescue for UK companies. Senior contractor who fixes, finishes and hardens apps built with Claude, Cursor, Lovable, v0 and friends. Production-ready code from AI prototypes — without throwing away what already works.
10+ years shipping production software · Lead Developer at Cap Energy · daily Claude Code user
Almost every "AI built our app in a weekend" story hits the same wall. Here is what I get called in to fix.
Edge cases, concurrent writes, weird inputs — the moment real traffic arrives, things start falling over.
AI tools love the visible 80%. The boring critical plumbing — sessions, billing, transactional email, audit trails — is missing or fragile.
No CI, no migrations, secrets in the repo, one server that nobody dares touch. A single bad PR away from a long evening.
The AI wrote it, the founder pasted it in, and now there is a 12,000-line codebase with no shared mental model behind it.
Pick the one that fits where your project actually is. Most engagements start with the audit.
Fixed-fee, ~1 week. A senior pair of eyes on your AI-generated codebase.
You inherited a vibe-coded prototype and need to know what you're actually shipping.
Take your AI prototype the last mile to a real product.
The demo wows on your laptop but breaks the moment a real user touches it.
Make a working app safe to leave running unattended.
It's live, it's earning, and you're lying awake worrying about it.
A senior engineer in the room with your AI tools.
You want to keep moving at AI speed without the technical debt catching up.
I'm a senior developer based in Oxford. I spend my days shipping production software — and using the same AI tools your prototype was built with. That gives me a useful double perspective: I know why AI-generated code looks the way it does, and I know what production traffic does to it.
Read more on my personal site: nathancollins.dev.
30 minutes, free. You show me the codebase and the pain. I tell you honestly whether I can help and what shape the engagement should take.
Fixed-fee deep read of the code, infra and dependencies. You get a written punch-list and effort estimates you can act on with or without me.
Based on the audit, a clear quote with milestones, deliverables and a delivery date. No timesheets, no surprise invoices.
I do the work, ship it to production, and hand over documentation so you (or your team) can keep moving.
Vibe coding is what people now call building software by describing what they want to an AI (Claude, Cursor, Lovable, v0, ChatGPT) and shipping whatever it generates. It is brilliant for prototypes and brutal for production: the code often works on the demo and falls over the moment real users, real data, or real edge cases turn up.
Yes — that is exactly the work. I use the same tools daily, so I can read AI-generated code quickly, spot the patterns that will bite you in production, and rewrite the parts that need rewriting without throwing away the parts that work.
I am based in Oxford and work primarily with UK-based companies — same time zone, UK contract law, UK invoicing in GBP. I will take occasional EU work where the timing is sensible. US clients welcome for short async engagements.
Audits usually start within 1–2 weeks of the discovery call. For genuine "live and on fire" situations I will try to clear the next available slot — get in touch and tell me how urgent it is.
No. Every engagement ends with documentation and a handover so that you, your team, or another contractor can pick up where I left off. The point of the rescue is to leave you with a codebase you control.
TypeScript, Vue/Nuxt and Node on the front and middle; C#/.NET and F# on the backend; PostgreSQL and SQL Server for data; Azure and Docker for infrastructure. I am stack-pragmatic — if your AI tool generated a React + Next.js + Supabase app, I can work in that too.
Often, yes — and I will tell you so on the discovery call if that is what I think. The trap is that AI is excellent at generating code and merely competent at understanding why existing code is broken. Once a codebase has accumulated a few rounds of AI patches on top of AI patches, an experienced human eye is usually what unblocks it.
Discovery calls are 30 minutes and free. If you'd rather write it out, the form goes straight to my inbox.
Pick a slot that works for you. UK time.
Embed not loading? Open Cal.com directly.
Reply usually inside one working day.