About David Ndungu

I'm a software engineer with 20+ years building software, including eight years at Zendesk across test, senior, and staff engineering roles. I have also run a development consultancy and founded an e-commerce SaaS that reached 50 paying retailers. Today I build open-source agent infrastructure and help engineering teams use coding agents effectively. I also build Sire.

What I actually build

My work spans verification with Kazi, decision history with Dira, agent memory with Serenity (in active development), and ML infrastructure with Zerfoo. Building and using these tools gives me concrete problems to work through, from agent handoffs to model execution.

What I do for clients

I help developers and engineering teams improve the workflow around Claude Code and other coding agents: acceptance checks, context, integrations, and handoffs. We start with one concrete bottleneck, agree on the scope and fee, and implement a result your team can evaluate and maintain. See the engagement and the engineering behind it.

At Zendesk

At Zendesk, I designed and delivered a zero-downtime etcd v2-to-v3 migration across production Kubernetes clusters and led diagnostic endpoint integration across 95% of production services. I worked on cloud-cost optimization through Kubernetes Cluster Autoscaler, production incident response, and service stress testing. As a Staff engineer, I delivered inbound email authentication with ARC and advocated gRPC adoption for SMTP Relay.

Building for customers

Before Zendesk, I founded and ran DavidDesigns, delivering custom websites and web applications. I later founded Gereji and launched an e-commerce SaaS for retailers, growing it from zero to 50 paying retailers and joining the Savannah Fund accelerator. That experience shapes how I scope consulting: a useful result, clear acceptance, and a handover the customer can maintain.

How that experience applies now

Production diagnostics and testing inform how I verify agent output. Cross-team infrastructure work informs how I capture decisions and coordinate handoffs. Cloud-cost work informs how I measure the cost of an accepted change, including retries and review. My current Fanisi work explores cheaper code generation; it is early experimental work, without a general savings claim.

Get in touch

The fastest way to reach me is the discovery call, or see the contact page for phone, email, and GitHub.