A patient picks their plan, uploads a prescription or types what they need, and the app reads it with an open-weight model, pulls the matching rules from that plan’s policy, and answers in plain language: likely covered, needs prior authorization, or within the plan’s limit. Suppliers, prices, stock and network membership all come from the database, never from the model, and every figure is labelled as sample data. A four-check match score shows exactly what’s settled and what still needs doing.