Production systems under real constraints.

The Problem

Healthcare scheduling is a constraint-satisfaction problem disguised as a calendar. Shifts have hard rules (certifications, labor law, rest periods) and soft preferences (seniority, fairness). Most tools treat it as a simple CRUD calendar. It isn’t.

Why It Was Hard

The scheduling engine evaluates hundreds of constraints per assignment and must return conflict feedback fast enough for interactive use. Each tenant has isolated data, its own rule configuration, and its own permission hierarchy. The backend carries the entire domain complexity — there is no "thin API over a database" shortcut here.

Architecture

• Constraint-based scheduling engine with configurable rule sets per tenant
• Multi-tenant data isolation at the database level (row-level security + tenant context)
• Role-based access control with hierarchical permissions
• Real-time conflict detection API that evaluates constraints on every assignment
• Event-driven audit trail for compliance and debugging

Key Decisions

• Chose PostgreSQL with row-level security over schema-per-tenant for operational simplicity at current scale
• Built the constraint engine as a pure function layer — no side effects, fully testable in isolation
• Separated the scheduling domain logic from the HTTP layer to allow future CLI/worker usage
• Used optimistic locking for concurrent schedule edits rather than pessimistic locks that would block the UI

Outcome

A working multi-tenant scheduling platform that handles real healthcare shift constraints — certifications, rest periods, labor rules, fairness preferences. Deployed and operational.