Architect of working systems.
I am Kiragu Maina, a systems architect based in Nairobi. I design and build AI-native enterprise systems, the kind that have to keep working through provider outages, rate limits, and 03:00 traffic spikes.
My favourite work lives in the layer most portfolios skip: queues, fallbacks, idempotency, observability, cost-aware routing, the audit trail. I treat the "hidden 90 percent" as the product.
I currently run Shellwire, an autonomous newsroom orchestrated by twelve LLM providers, and I take selected contract engagements with teams that need an architect, not another tutorial.
I write the architecture decision before the first commit.
Trade-offs, constraints, the failure mode I am designing for. If I cannot defend the diagram, the code does not deserve to exist.
I prefer Postgres until proven otherwise.
Row-level security, materialised views, JSONB, listen/notify. Most teams reach for a new database before they have used the one they have.
I plan for the provider to go dark.
Three-tier fallback chains, ranked-fallback orchestration, cached last-known-good. AI vendors fail. Networks fail. Pipelines should not.
I instrument before I optimise.
Latency, tokens, queue depth, error rate, by provider. You cannot tune a system you cannot see.
I keep the dev loop tight.
Docker Compose for local parity, hot reload, fast tests, real seed data. Friction in the loop is the largest hidden cost in a project.
