I'm the engineer you call when the AI demo has to become a product.
The software your business runs on: the parts customers touch, the parts they never see, and the AI in between.
It usually starts one of two ways. Something broke and your customers noticed, or something is about to go live and has to be right the first time.
Systems that were not allowed to fail.
6 shipped between 2023 and 2026. Every figure below is measured against a stated baseline and explained in the writeup.
- 2026
WhatsApp CRM
Leads die in an inbox nobody owns.
one shared inboxtraced handoffs - 2026
AI underwriting assistant
Credit decisions on real applicants.
adopted in 10 weeks120-person team - 2025
Inference routing
One provider outage becomes a product outage.
28% lower spendp95 sub-second - 2025
LLM eval harness
A silent regression ships to every user.
18% higher pass rate - 2024
Fraud detection
Every miss is a chargeback someone eats.
70% less fraud55% less review load - 2023
Payment settlement
Settlement halts and money stops moving.
99.999% uptime20M+ txns / month
Something in production has to hold.
Workflow automation, AI products that have to become real, fintech systems, custom backends, architecture review, and senior engineering leadership by the engagement.
Scale-ups, funded startups, professional firmsFor small businessYour business needs to be online and answering.
WhatsApp AI customer service in English and Swahili, professional online setup, M-Pesa wired correctly, emergency website rescue, and AI-built apps checked before real customers touch them.
Solos, micro-businesses, small practicesThe person who scopes your work is the person who builds it. Clients engage me directly or through Noria Technologies LTD, the practice I have run since 2020.
The notes cover the decisions founders hit before they hire anyone: what AI actually costs per user, what to automate first, and why an AI-built app is a prototype until production proves otherwise.