EcomAI — AI-Powered E-Commerce Automation Platform

Multi-tenant backend for AI-driven social commerce: auto-replies on WhatsApp, Instagram & Facebook DMs using Meta Webhooks + Gemini/OpenAI, with an autonomous marketing engine powered by Claude + RAG.

---

What's Implemented

✅ Core Messaging Automation

• Meta Webhook processing — WhatsApp, Instagram DM, Facebook DM & comments
• AI-powered replies — Gemini (default) or OpenAI, switchable from platform settings
• Inventory-aware responses — AI reads product catalog before replying
• Conversation threading — per tenant × platform × customer
• Comment handling — stored only (no auto-reply), DMs get full AI response

✅ Multi-Tenant Architecture

• Host/tenant separation with IsHost flag
• Tenant provisioning (create, suspend, activate) from host admin
• Scoped EF global query filters per tenant
• RBAC — roles, permissions, user-role assignment, dynamic JWT claims

✅ Meta OAuth Integration

• OAuth connect/callback/disconnect per tenant
• Multi-asset discovery: Facebook Pages, Instagram accounts, WABAs, phone numbers
• Page webhook subscription post-OAuth
• Encrypted token storage at rest (ASP.NET Core Data Protection)
• DB-backed platform Meta config (AppId, AppSecret, GraphVersion, CallbackBaseUrl)

✅ Frontend (Angular)

• Auth with host/tenant toggle, JWT session restore
• Tenant management screen (host-only)
• Messaging inbox — real-time capable, split-panel, mobile responsive
• Platform settings screen
• OAuth callback handler
• Webhook tester (dev tool)

✅ Autonomous Marketing Engine (Architecture Complete)

• Full spec and DB schema for Claude-powered ad decision engine
• RAG foundation with pgvector for knowledge retrieval
• 3-phase rollout: System Prompt → Knowledge Base → Full RAG Memory
• Skill-file-driven system prompt (no redeployment to change strategy)
• See Marketing_Engine_Plan_v3.docx for complete spec

---

Tech Stack

Layer	Technology
Runtime	.NET 10 / ASP.NET Core
ORM	EF Core 10 (SQL Server)
Messaging	MediatR 12
Background Jobs	Hangfire 1.8
Resilience	Polly 8
Logging	Serilog
Realtime	SignalR
AI Providers	Gemini, OpenAI, Ollama (switchable), Claude
Frontend	Angular (SPA) + PrimeNG
Auth	JWT + refresh token rotation + RBAC

---

Project Structure

AI-Whatsapp/
├── Ai-Business/          # Domain entities, commands, interfaces
├── AI-Infrastructure/    # EF Core, repositories, external services, auth
├── AI-Whatsapp/          # API host, controllers, middleware, composition root
├── tests/UnitTests/      # xUnit unit tests
docs/
├── domain_knowledge.docx         # Full implementation documentation
├── Marketing_Engine_Plan_v3.docx # Autonomous marketing engine spec

---

Getting Started

Prerequisites

• .NET SDK 10.0.103
• SQL Server (or Azure SQL)
• Meta App with WhatsApp/Instagram/Facebook products configured

Configuration

Copy and fill environment variables / user secrets:

{
  "ConnectionStrings": {
    "DefaultConnection": "<sql-server-connection-string>"
  },
  "JwtSettings": {
    "SecretKey": "<min-32-char-secret>",
    "Issuer": "EcomAI",
    "Audience": "EcomAI",
    "AccessTokenExpiryMinutes": 60,
    "RefreshTokenExpiryDays": 30
  },
  "MetaOAuthSettings": {
    "AppId": "<meta-app-id>",
    "AppSecret": "<meta-app-secret>",
    "GraphVersion": "v21.0",
    "CallbackBaseUrl": "https://yourdomain.com"
  },
  "MetaWebhook": {
    "VerifyToken": "<your-webhook-verify-token>"
  },
  "AIProvider": {
    "ActiveProvider": "Gemini",
    "Gemini": { "ApiKey": "<gemini-api-key>", "Model": "gemini-1.5-pro" },
    "OpenAI": { "ApiKey": "<openai-api-key>", "Model": "gpt-4o" },
    "Ollama": { "BaseUrl": "http://localhost:11434", "Model": "llama3" }
  }
}

Production: Store secrets in environment variables or a secrets manager. Never commit to source control.

Run Migrations

cd AI-Whatsapp
dotnet ef database update --project AI-Infrastructure --startup-project AI-Whatsapp

Run

dotnet run --project AI-Whatsapp

Swagger available at https://localhost:{port}/swagger in development.

---

AI Provider Switching

The active AI provider is configurable at runtime via Platform Settings (/host/platform in the UI or PUT /api/host/platform/meta). No redeployment needed.

Supported providers: Gemini | OpenAI | Ollama | Mock (zero-cost debug)

---

Meta Webhook Setup

1. In Meta Developer Console, set webhook URL to: https://yourdomain.com/api/webhooks/meta
2. Set verify token to match MetaWebhook:VerifyToken in your config
3. Subscribe to: messages, messaging_postbacks, feed, mention
4. Connect tenant channels via the Integrations screen in the UI

---

Marketing Engine (Pending Implementation)

The autonomous marketing engine (Claude + Meta Marketing API + pgvector RAG) is fully architected and documented but not yet implemented in code. The spec covers:

• Daily Hangfire agent job (analyze → decide → execute on Meta Ads)
• pgvector RAG for knowledge retrieval and decision memory
• Safety controls: DryRun mode, spend caps, approval gates, rollback
• Gradual rollout plan (Advisory → Supervised → Partial → Full Autonomy)
• Gemini Veo creative pipeline

See docs/Marketing_Engine_Plan_v3.docx for the complete spec. Contributions welcome.

---

Pending / Known TODOs

• Complete Meta outbound methods (template, image, quick-replies, read receipts)
• Real email provider for password reset (currently logs token)
• Idempotency for duplicate webhook events (by ExternalMessageId)
• Auth hardening: JWT key to secrets store, CORS restricted to known domains
• Remove legacy ClientSecretsRepository WhatsApp fallback in WebhooksController
• Connect Messaging inbox to SignalR for live message delivery
• Angular RBAC management screen (/admin/rbac)
• Angular screens: Content AI, Products, Scheduling (currently placeholders)
• Implement Marketing Engine services (see spec)
• Integration tests for webhook signature, tenant resolution, policy enforcement

---

Contributing

This project is open source. Contributions are welcome across any layer — backend services, Angular screens, test coverage, or the marketing engine implementation.

Please open an issue before large PRs to align on approach.

---