Built for Scale & Safety
Explore the complete system architecture powering AeroGuardian AI — from user interface to ML inference pipeline.
User Workflow
End-to-end journey from authentication to emergency response.
Authentication
Users sign up / log in via Supabase Auth with email or OAuth providers.
Dashboard Access
Authenticated users access the real-time flight risk monitoring dashboard.
Data Streaming
Simulated flight telemetry is streamed to the dashboard via API polling.
AI Predictions
ML models analyze telemetry and return risk scores & anomaly detection results.
Alert & Response
High-risk flights trigger alerts with emergency airport recommendations.
Action & Review
Operators review AI insights, approve diversions, and monitor resolution.
Technical Architecture
Four-layer architecture with clear separation of concerns and scalable microservice design.
Presentation Layer
Next.js 14 (App Router)
Server & client rendering
React + TypeScript
Type-safe UI components
TailwindCSS + ShadCN
Design system
Framer Motion
Animations & transitions
Recharts
Data visualization
API Layer
/api/predict-risk
AI risk prediction endpoint
/api/telemetry
Telemetry data generation
/api/anomaly-detect
Anomaly detection pipeline
/api/emergency-recommendation
Emergency routing
AI / ML Engine
FastAPI Microservice
Python ML serving
Scikit-learn
Model training pipeline
XGBoost / RandomForest
Risk prediction models
Anomaly Detection
Statistical thresholding
Data Layer
PostgreSQL (Supabase)
Persistent data store
Supabase Auth
Authentication & user management
Telemetry Pipeline
Simulated flight data
Airport Database
Emergency diversion data
AI Engine
End-to-end machine learning pipeline from raw telemetry to actionable safety recommendations.
Data Ingestion
Raw telemetry arrives: engine_temp, altitude, fuel_usage, weather_risk, vibration, oil_pressure
Feature Engineering
Normalize, compute rolling averages, detect trend changes, calculate rate-of-change features
Model Inference
RandomForest ensemble (100 trees) + XGBoost gradient boosting → weighted ensemble prediction
Post-processing
Risk classification (Low/Medium/High/Critical), confidence scoring, anomaly flagging
Action Engine
If risk > threshold → compute nearest airports, generate diversion route, emergency procedures
Model Specifications
Infrastructure
Production-ready deployment architecture with zero-cost hosting options.
Vercel
Frontend hosting with edge network, automatic CI/CD, and serverless functions.
Render
Python ML microservice deployment with auto-scaling and health checks.
Supabase
PostgreSQL database + Auth + Real-time subscriptions on free tier.
Security
HTTPS, JWT tokens, row-level security, CORS protection, input validation.
CI/CD
Git-based deployments, preview environments, automated testing pipeline.
Performance
Edge caching, server components, streaming SSR, optimized bundle splitting.