The first direction was an AI co-pilot for Community Health Workers. It would have been another surface in an already fragmented workflow. At validation stage the risk was familiar: launch, low adoption, blame change management, never surface that the wedge was wrong.
Field contact with CHWs, ACPs, and PCPs ran before significant build, not to defend the concept but to locate where time, trust, and context were actually lost.
Outcomes.
Direction changed pre-build
Co-pilot abandoned; Lorena scoped upstream to voice outreach and structured handoffs before human follow-up.
Burden removal wedge
Workflow-first voice + RAG tied to the shifts CHWs actually work.
Reusable infrastructure
Patterns carried across healthcare incubations at Red Cell.
What the full case covers
- Field research process and findings that changed the roadmap
- Interaction and voice architecture for Lorena
- Human-in-the-loop escalation and accountability model
- Technical architecture and incubation reuse