The situation
The Cairya coaching platform had demonstrated early commercial viability with negotiation and leadership clients. The next question was whether AI-assisted intake could extend into more sensitive territory — mental health and psychosocial assessment — while remaining safe to deploy. This is a fundamentally different design problem than general coaching: the stakes of a wrong response are clinical, not just experiential.
Provider organizations needed a tool that could efficiently pre-screen clients using standardized psychosocial assessments, surface risk signals early, and hand off to a human provider at the right moment. The design challenge was figuring out exactly where AI could operate reliably, where it needed to pause and prompt for human review, and where it had to refuse and escalate immediately — before the session started, not after something went wrong.
Solving the problem
The core design artifact was a swimlane journey map covering three actors: the patient, the provider, and the AI system ("Ana"). For each step in the psychosocial assessment flow, the map documented which actor owned the action, what capability was available (AI available, workaround needed, or function not yet implemented), and what safety constraints applied.
- Defined a three-tier safety escalation model. Tier 1 covered normal AI-assisted intake — structured screening, standardized questions, data capture. Tier 2 covered flagged responses — moderate risk indicators that prompted Ana to pause, summarize context for the provider, and transfer control. Tier 3 covered crisis indicators — immediate escalation to a live provider with no AI mediation, and a hardcoded routing rule that blocked Ana from continuing the session regardless of prompt framing.
- Documented capability status per interaction step. Every action in the journey was tagged: Available (AI can handle this now), Workaround (current implementation partially handles it), or N/A (not yet implemented, human must cover). This gave the engineering team a design-driven feature backlog and gave the clinical advisor a clear view of where the system was and wasn't safe to deploy.
- Made HIPAA constraints visible in the design artifacts. Rather than treating data handling as a backend concern, interaction patterns that touched PHI — intake responses, risk assessments, session notes — were annotated in the journey map with handling requirements. Design decisions about what Ana could store, repeat, or summarize were made explicitly, not left to implementation defaults.
- Delivered a working MVP prototype validated by clinical review. The prototype at cairya-providerportal-mvp1.netlify.app was reviewed against the journey map by a clinical advisor before release, confirming that escalation triggers fired correctly and that no Tier 3 indicator could be completed by Ana without live-provider handoff.