RiveHub — Architecture

RiveHub is a production, multi-tenant SaaS that automates MAPAQ regulatory compliance for Quebec restaurants. I designed, built and operate it solo.

The product source is private (commercial). This repository is a high-level system-design writeup, plus the AI-assisted development methodology behind it.

Product: rivehub.com

What it does

Quebec restaurants must satisfy MAPAQ food-safety rules (HACCP logs, traceability, temperature records, labelling). RiveHub turns that manual, paper-driven burden into automation:

• AI document intake : receipt & invoice OCR/extraction; menu extraction from URLs or files
• HACCP automation : builder, checklists, photo capture, audit-ready records
• 24/7 IoT temperature monitoring : sensor ingestion plus threshold alerts
• Operational AI : prep-list generation, demand forecasting, menu engineering, best-before labels, content generation
• i18n : 59 locale catalogs for multilingual kitchen brigades

System architecture

flowchart TD
  U[Restaurant staff: web and mobile] -->|Next.js 16 App Router| APP[Next.js / React 19]
  APP -->|REST / JSON| API[40+ API routes / 20+ AI-integrated]
  API --> SB[(Supabase: PostgreSQL + RLS / 60+ tables, multi-tenant)]
  API --> CLAUDE[Anthropic Claude API: OCR, extraction, generation, analysis]
  API --> STRIPE[Stripe: billing]
  IOT[IoT temperature sensors] -->|ingest| API
  CRON[Scheduled jobs: prep lists, forecasts] --> API
  APP -.deployed on.-> VERCEL[Vercel]

Loading

• Multi-tenant isolation enforced at the data layer with PostgreSQL Row-Level Security across 60+ tables, so every query is tenant-scoped by default.
• 40+ API routes, of which 20+ are AI-integrated (document extraction, menu analysis, forecasting, content generation, notes analysis).

AI subsystem (production, not demos)

• Structured extraction : receipts, invoices and menus parsed into typed, validated records (structured outputs / tool use).
• Cost and latency control : prompt caching and batching where the workload allows.
• Safety : input validation and prompt-injection defense on user-facing AI endpoints (kept high-level by design).
• Provider strategy : Anthropic Claude primary, with graceful degradation.

Built by an agent fleet

Developed and operated with a fleet of Claude Code agents I built, an orchestration harness that turns one developer into a team:

flowchart LR
  O[Orchestrator: decompose, delegate] --> E[Executor: implement]
  E --> V[Verifier: test, audit]
  V -->|pass| S[Ship: build, deploy, verify]
  V -->|fail| R[Repair: diagnose, fix]
  R --> V

Loading

Topology patterns in rotation: Solver + Verifier, Pipeline, Audit Fleet, covering development, QA, deployment and incident response under a contract-based coordination protocol.

Stack

Layer	Tech
Frontend	Next.js 16 (App Router), React 19, TypeScript
Data	Supabase (PostgreSQL + Row-Level Security), 60+ tables
AI	Anthropic Claude API, MCP, multi-agent orchestration
Payments	Stripe
i18n	next-intl (59 locales)
Infra	Vercel, CI/CD, Git

---

By Nassim Saighi, MD, physician + applied-AI builder. Source private; happy to walk through the system on request.