I build production-grade AI systems.

LimoHive books thousands of rides a week, and every call that went unanswered was a deal walking out the door. I built an end-to-end agent that plugs into their CRM and booking flow, takes over the moment a customer starts a reservation, and calls them within seconds to confirm pickup, drop off, airport transfers, and round trips. It runs 24/7 with AI voice, follows up on leads that go quiet, and hands the conversation to a human whenever a deal needs one.

The agent lifted closed deals and payments by more than 60% compared to the manual process it replaced, and gave the sales team back their day. Instead of dialing cold leads, they now spend their hours on high-value conversations, and every lead gets the same quality of follow-up whether it arrives at noon or 2am.

Codex and Claude are excellent tools, and the bill for a full engineering team is brutal. I built the company's own coding agent from the ground up: a custom harness built on open source coding agent strategies, routed through a dynamic mix of open models so every task goes to the model that handles it best. Under it sits a mixed compute cluster I set up for agent workloads, with NVIDIA nodes for heavy inference, macOS servers for platform-specific builds, and dedicated machines running QA review agents that audit code and live websites in the background.

The economics are the headline. What would cost a 30-person engineering team about $7,000 a month in commercial coding seats runs on roughly $100 a month on our stack, a saving of about 70x, and every engineer at DIQ now writes, reviews, and ships with it daily. The same cluster also deploys our client agents, so one build powers both our internal speed and the systems we deliver.

A top US PEO firm was running its entire social presence by hand, across their website, blog, Instagram, Facebook, and X. Posting was inconsistent, and every quiet week cost them reach. I designed a social engine where autonomous agents run the full pipeline: trend research, content drafting, scheduling, publishing, and reporting. The agents sit on a hybrid harness that combines Hermes with an orchestration layer we built in house, and every stage runs on a model chosen for that specific job, which is what makes the output fast, on-brand, and dependable.

The engine has driven millions of views across their channels and compounded their growth: page visits climbed, SEO ranking improved, and the brand now publishes every single day, weekends included. Their team moved from chasing posts to reviewing results, and the reach keeps building on its own.

Payments in Africa do not look the same from one country to the next. Each of M-Kopa's five markets, South Africa, Nigeria, Ghana, Kenya and Uganda, runs on its own rails, rules and regulators, and a checkout that works in one can fail in the next. I travelled across the markets to learn each one properly, from local payment policies and PCI DSS compliance to encryption standards, then led the Android team as we folded all of it into a single interface. The app now adapts to the country a customer is paying from, so the experience feels local while the team maintains one product instead of five.

The platform handles daily loan recollection on the phones and assets M-Kopa finances, and accepts payments in every form each market needs, all over a secure and compliant stack. It became the payment backbone for the business across all five countries.

M-Kopa puts smartphones in the hands of low income earners on credit, and that only works if the device can be locked the moment a payment is missed. Software-only locks are easy to bypass, so I worked with Samsung and Nokia to push the locking into the hardware itself, using Nokia's soft lock and the latest Android security SDKs so the firmware cannot be flashed, reset or worked around.

The result is a remote locking process that holds up in the field. When a payment is due, the device locks reliably and the loan stays protected, which is what lets M-Kopa keep extending credit to people who would otherwise be turned away.

Designed and scaled payment products on Android and iOS, and introduced test-driven development that reduced software bugs by 60 percent.

Led the launch of a merchant app that reached more than 10,000 daily active users within two months, with documentation and testing standards the team kept using.

Built inventory and payment-collection tooling that replaced white-label dashboards, with cloud workflows that automated back-office operations end to end.