Banks & enterprise SaaS
Multi-tenant platforms
One codebase now serves Wema Bank, First Bank, PremiumTrust Bank and Seven-Up, with over 10,000 production users, where each client used to run its own copy. Lawyering, built solo with agents, has been in production since June 2026: 19 modules, over 1,200 tests.
Goal: Sell one product to many organisations, each with its own branding, rules and infrastructure, without keeping a codebase per client.
Lawyering gives law firms and individual practitioners each an isolated workspace for their staff and clients. Tangerine365 separates tenants by subdomain or by ID. At Seamless I work on tenant isolation for a multi-tenant HR suite.
Solutions architecture
Deployment and infrastructure design
Cut the hosting bill for Zercom's internal tools, staging and UAT by about 85%, and wrote the deployment architecture for First Bank's three-tier on-premise build.
Goal: Run each system on infrastructure that fits its budget, its compliance rules and its risk.
For client deployments I write the sizing, server requirements and deployment architecture. For Zercom's own environments I moved separate DigitalOcean droplets onto one server running CyberPanel and LiteSpeed, configured for wildcard subdomains.
National health campaigns
Offline-first data systems
Traced field-worker lockouts to activities being dated at sync time instead of creation time. Now sizing infrastructure and support staffing for a 2027–29 programme across 11 states (~27M people).
Goal: Field workers in malaria campaigns get credited and paid for the days they actually worked, even when their phones sync late.
On contract with eGov Foundation I support DIGIT HCM, the platform behind Nigeria's health campaigns, across Abuja, Bauchi, Taraba and Borno.
Agent governance
Guardrails for AI coding agents
Three open source npm packages. In my experiment, agents asked to purge old orders hard-deleted them 4 times out of 4; given a reviewed decision record, they kept the order numbers 3 times out of 3.
Goal: Get the speed of AI coding agents on production codebases without losing control of what they change.
One package for each point where an agent's work can go wrong. agent-contract blocks any write outside the task's scope before it lands. repo-memory keeps the reasons behind the code as records that CI checks and only a human can accept. change-evidence reads every pull request as data and tells the reviewer where to look. I build Lawyering under all three.
