Architecture & Two-Stage AI Scoring
InBox-AleRt is designed to filter out email noise and alert professionals instantly via WhatsApp when critical communications arrive. It uses a high-throughput webhook ingestion layer coupled with asynchronous AI scoring.
Engineering Pipeline Architecture:
[Gmail Pub/Sub / Outlook Graph API Webhooks] ──► [FastAPI Router]
│
▼
[Celery Task Queue & Redis]
│
▼
[Stage 1: Rule & Sender Filter]
│
▼
[Stage 2: LLM Scoring Engine]
• Evaluates priority (0-100)
• Generates 2-sentence summary
│
(Score >= Threshold)
▼
[WhatsApp Business API Dispatch]
Key Technical Features
- Two-Stage AI Scoring: Stage 1 fast rule evaluation + Stage 2 LLM semantic analysis scoring emails from 0 to 100 with concise executive summaries.
- Real-Time Webhook Subscriptions: Real-time event push handlers for Google Cloud Gmail Pub/Sub and Microsoft Outlook Graph API.
- Enterprise Security: OAuth 2.0 authentication, RS256 signed JWT tokens, and AES-256 encryption for access token storage at rest.
- Containerized Stack: Orchestrated with Docker and Docker Compose across Next.js frontend, FastAPI REST backend, Celery workers, and Redis broker.