Architecture decisions, deployment lessons, and engineering insight from building AI agents, SaaS platforms, and operational tools — published as they are written, not when they are polished.
A walkthrough of how the Lead Agent Control Plane was designed — from the decision to build a centralized orchestration layer instead of deploying individual agents site-by-site, to the routing logic, the conversational context management, and what broke in staging that we didn't see coming.
The data model decisions, multi-tenant structure, and workflow engine behind ManuTrack — and why the constraints of the West African market shaped the architecture in ways a standard playbook wouldn't have.
How to embed AI into WordPress without treating it as a plugin add-on — covering custom REST endpoints, streaming responses, context injection from post data, and managing latency in shared hosting environments.
The patterns we kept and the ones we abandoned after running AI agents in live business workflows. System prompt design, context window management, output formatting contracts, and why chain-of-thought isn't always the right call.
The technical and design decisions behind Sahihi's verification architecture — from the data flow of a verification request to the UX of communicating trust states to end users, and what the West African identity verification landscape actually looks like.
The internal framework for how agents are designed, deployed, and coordinated across the Tamensah Labs ecosystem — covering role definition, context routing, failure handling, and the interface between agents and the systems they touch.
What changes when your users are on mobile networks with variable bandwidth, your payment infrastructure is mobile money, and your competition is WhatsApp. A ground-level view of building in Ghana and what it teaches you about software that actually works.
A complete walkthrough of how SolidStartup AI went from a concept to a deployed product — covering the AI workflow design, the data model, the decisions on what to build vs. buy, and the three things we rebuilt before launch.
Running AI-powered applications where internet access is intermittent changes every assumption about latency, caching, and fallback logic. What we learned deploying AI tools in environments where the network is not reliable infrastructure but a variable resource.