Skip to content

architecture.md

Latest commit

 

History

History
106 lines (80 loc) · 1.94 KB

architecture.md

File metadata and controls

106 lines (80 loc) · 1.94 KB
Raw

🏗 Architecture – AI Multi-Channel Outreach Platform

1. Architectural Philosophy

The system is designed as:

  • Multi-tenant SaaS
  • Event-driven automation engine
  • Async-first processing
  • AI-native personalization layer
  • Horizontally scalable

2. High-Level Architecture

Frontend (Stitch MCP HTML/CSS/JS) ↓ Django API Layer (Business Logic + Multi-Tenancy) ↓ PostgreSQL (Primary Structured Data) ↓ Firebase

  • Realtime notifications
  • Live campaign dashboards
  • Websocket-like UI sync ↓ Task Queue (Celery Workers) ↓ Messaging Providers
  • Email
  • SMS
  • WhatsApp
  • LinkedIn Automation Layer ↓ Gemini AI Service (Content Engine)

3. Core Architectural Layers

A. Presentation Layer

  • Stitch MCP generates HTML templates.
  • Tailwind-based UI.
  • AJAX/Fetch for API communication.
  • Firebase used for live dashboard updates.

B. Application Layer (Django Apps)

apps/ │ ├── tenants/ ├── users/ ├── leads/ ├── campaigns/ ├── sequences/ ├── dispatch/ ├── ai_engine/ ├── analytics/ ├── compliance/ └── monitoring/

C. Automation Engine

  • CampaignLead acts as state machine.
  • Celery schedules next_execution_time.
  • Workers execute steps.
  • Conditional branching handled server-side.

D. AI Layer

AI abstraction layer:

  • generate_email()
  • generate_sms()
  • generate_linkedin_message()
  • generate_subject_line()
  • analyze_reply_intent()

All prompts routed via Gemini wrapper service.

Provider-agnostic design.

4. Multi-Tenant Isolation

  • Shared database
  • Strict organization_id on root tables
  • Custom TenantMiddleware
  • Automatic ORM filtering
  • Hard database-level unique constraints per tenant

5. Scalability Principles

  • Stateless Django web nodes
  • Separate worker pool
  • Read-optimized analytics snapshots
  • GIN indexes on JSONB fields
  • Queue partitioning by job type