The interesting problem was never making a model talk.
It’s making it reliable enough for a business to trust with real customers. That’s the standard I hold every system to — because I spent four years in industries where software failing meant money moving wrong.
I started as a full-stack engineer building the systems businesses actually run on — fintech platforms handling real transactions, ERP consolidations, point-of-sale systems wired into accounting and payments. Four years of learning what “production” really means: compliance audits, data that can’t be wrong, uptime people depend on.
When LLMs arrived, I watched everyone build impressive demos that fell apart on contact with real customers. The gap was obvious: the AI world had plenty of prototypes and very few engineers who knew how to ship. So I brought the production discipline to the AI side — agents at Nexnology that chased overdue invoices inside a PCI DSS-compliant fintech, analytics agents turning transaction data into decisions.
Now I do it independently: AI systems for businesses that need them to actually work, from Karachi, for clients anywhere.