Because the platform was envisioned as a scalable solution that could eventually serve other non-profits and healthcare providers, we built tenant management as the first layer. A super admin or system admin can create and configure tenants, define tenant-level settings (including whether assessment forms are MCP-specific or generic), assign tenant admins, and activate or deactivate tenants as needed.

Tenant admins inherit a focused subset of super admin capabilities, allowing each person on the platform to manage its own users, configurations, and data without affecting others. This separation also keeps client data isolated by tenant, which was important from a privacy and compliance standpoint.

Patient management was the heart of the platform. We split it into two clearly defined stages, owned by different roles, so that handoffs were explicit and accountability was clear. The intake coordinator is responsible for client onboarding. From their dashboard, they can launch an onboarding form that captures structured data across eight sections: client identity, demographics, contact information, current address, emergency contact, identifications, eligibility, and essential items.

More importantly, the form is paired with an in-app interaction recorder. The coordinator can start, pause, resume, and stop a recording of the intake conversation, with explicit consent captured before recording begins. This solved a long-standing problem in which intake conversations were either lost or summarized inconsistently.

Billing was the most operationally painful area for our client, and we focused the most engineering effort there. The platform’s billing module is designed to account for the fact that each MCP has its own rules. Assessment forms, eligible service codes, documentation requirements, and billing cadences differ between Kaiser, Health Net, and PHC. We built a configurable billing engine that defines these rules per MCP and applies them automatically when claims are generated.

Two AI capabilities were core to the platform: transcription of recorded client interactions and assistance with care plan generation. Both were initially scoped against third-party APIs, but at the volumes the organization was projecting, third-party costs would have grown into a meaningful operating expense over time.

We proposed a self-hosted approach. For speech-to-text, we deployed Whisper on the client’s infrastructure, enabling the team to obtain accurate transcription of intake and assessment conversations without per-minute API charges. For care plan drafting, we integrated Gemini into a controlled flow in which care managers receive a structured, suggested plan based on assessment data, which they can then review, edit, and finalize. The combination kept clinical control firmly in the care manager’s hands while removing the blank-page problem that slows documentation. Hosting these models internally also strengthened the privacy posture, as recordings and assessment data did not have to leave the client’s environment for processing.

Before the platform, generating any kind of report meant pulling data from spreadsheets and shared drives by hand. We built a reporting layer that gives leadership real-time visibility into the metrics that matter most: number of clients onboarded, clients in assessment, active care plans, claims submitted, claims approved, claims rejected, and revenue by MCP. The reporting layer ships with a baseline set of dashboards and supports customized views, where users can select the data and metrics relevant to a specific funder report or internal review.

This was important for the non-profit context, where reporting requirements differ by funder and grant.