Skip to main content

Architecture Overview

The Problem

Korea’s social service industry — child rehabilitation, welfare, education, pharmacy — still runs on fragmented legacy systems. The core problem isn’t just “digital transformation.” It’s the absence of an architecture that can answer: “Who verified an AI decision, how, and why?”

The Solution

Cronozen is a full-stack platform with Decision Proof Unit (DPU) as its core primitive. Three design principles:
  1. Provable Decisions — Every AI recommendation is sealed in a DPU envelope with evidence, confidence scores, risk levels, and approval chains
  2. One Backend, Infinite Context — Rehab centers, welfare agencies, pharmacies all run on the same infrastructure with fully isolated tenant contexts
  3. Policy as Runtime — National, regional, and center-level policies are automatically applied at runtime with temporal validity

Hub-and-Spoke Ecosystem

OPS Hub

The central hub manages all cross-cutting concerns:
  • SSO — Single sign-on via auth.cronozen.com
  • Payments — Toss Payments, subscription management, billing
  • Tenant Master — Center/workspace provisioning and lifecycle
  • DPU — Decision Proof Unit with SHA-256 hash chains
  • 7 Verticals — Rehabilitation, welfare, education, pharmacy, commerce, mentoring, interior

Spoke Services

7 Verticals

All verticals share the same codebase, differentiated by tenant configuration: Each vertical uses Centers as the isolation boundary. Centers have workspaces, and within workspaces all data, permissions, and policies are scoped.

Tech Stack

Data Flow

Deployment Strategy

All repos follow the same pattern: develop → staging, main → production.

Security Architecture

  • Tenant Isolation: Row-level via center_id scoping in Prisma middleware
  • Auth: JWT with multi-tenant context, actor family linking
  • Audit: Append-only logs, 12 event types, basePrisma usage audited
  • Cron: All 39 cron routes require verifyCronSecret()
  • Data Access: All center access through center_memberships only — no backdoor
For deep dives into specific subsystems, see Multi-Tenant Architecture, DPU Engine, and Proof Pipeline.