Hi, I’m Xander, building AI systems for healthcare and community in Nairobi

I’m John Alexander Kamau, co-founder of Norah Labs, a small studio in Nairobi building AI products and software for problems close to home.
What I’ve shipped and am working on:
1.) Tabibu Health AI: a LangGraph agent for healthcare, with retrieval across three sources, guardrails on both input and output, and emergency routing.
2.) approvals-desk: a human-in-the-loop console for LLM-proposed refunds, built around exactly-once execution and an audit trail.
3.) Teleios Connect: a React Native community app with 200+ weekly users across 7 countries.

The thread through all of it is what I call Behavioral Systems Design: a system is only as good as the read on the people using it. The architecture matters, but most failures I’ve seen come from wrong assumptions about behavior.

Something I'm looking forward to from this group:

• If you’ve shipped AI in healthcare or another high-stakes domain, how do you test emergency routing and guardrails beyond unit tests?