Seven years building softwares that aren't allowed to fail. Two consumer brands — Abirash and XEYLA — that I run on my own money, where every system gets tested before it goes near your product.
Free · I reply within 24 hours · no deck, no retainer talk
Funded team, real users, and a prototype that impressed everyone until it met real traffic. You want an engineer who thinks in outcomes and P&L — architecture, cost, and failure modes on the table before a line gets written.
One or two outlets, everything living in WhatsApp and a notebook. You hear "AI" constantly and have no idea where to start. I'll point you at something already running in Abirash or XEYLA — no theory, no pilot.
Most of my seven-plus years as an engineer have gone into software that moves money. It's the kind of code that isn't allowed to fail quietly at 2am, and it leaves you with a particular sort of paranoia.
Then I opened Abirash — a sneaker label where I run operations and marketing end-to-end on AI: stock, shelf, ad creative, and the DMs that close the sale. The first six months taught me more about software than any year at a desk had — not because the work was harder, but because the feedback was instant. A bug gets logged and triaged. A customer who walks out just leaves.
That changed what I'm willing to ship. I stopped being impressed by things that work in a demo and started asking what happens on the worst day. XEYLA sharpened it further — an authentic Pakistani dress line where the same AI runs both the back office and the campaigns, because when authenticity is the entire product, what you automate has to be right before it's fast.
So nothing reaches a client that hasn't already cost me money first. It's the one credential I can't fake.
The moatAn AI engineer can build a demo. An operator can describe the real problem. I do both in the same week, on my own money.
Engineers don't carry inventory. Brand owners can't ship code. I'm the overlap.
Alpha Centuri is where client work lands. The two brands are the lab it gets tested in.
Where product companies hire me. Builds AI-native software for clients across the US, UK, and Europe — architecture, cost, and failure modes agreed before a line gets written.
alphacenturiai.com →Real inventory, real customers, real margins. Nothing reaches a client that hasn't already cost me money here first.
abirash.com →Genuine Pakistani dresses, imported and sold direct — no middleman markup, no guesswork on quality. Authenticity is the entire product, which makes it an honest test of what AI can and can't claim.
xelya.xyz →Most AI projects fail because nobody defined the expensive pain first. Demos are cheap; outcomes are the product.
The interesting engineering is all in the failure modes — rate limits, bad inputs, the model being confidently wrong at 2am. That's the part nobody demos.
Carrying inventory, chasing a customer, eating a loss makes you a dramatically better engineer — and a much harder person to sell a bad architecture to.
Order volume, conversion, and cost-per-result from Abirash and XEYLA land here as the experiments close out. I don't publish a stat I can't back up.
Tell me what it needs to do and where you think it breaks. I'll come back with an approach, the failure modes, and a rough cost — before either of us commits.
How I build software and business with AI, plus agentic and AI-engineering breakdowns from the builds themselves.
The plan is 40+ AI builds, each written up the same way: the setup, the result, and what it cost — including the ones that failed. Most run through Abirash and XEYLA first, because that's where I can publish real numbers without a client's permission.
One message. I'll come back with an approach and the failure modes I'd worry about — or tell you straight that I'm not the right person.