Architecture Visual

Complete production architecture for TrendDraw, from static delivery through serverless trend analysis, AI-assisted suggestions, and scheduled data ingestion.

Delivery and Frontend

User BrowserReact 19 SPA, dashboards and charts

→

CloudFrontShared with the main portfolio distribution

→

S3 Static Hostingjoesparkman.com/trend-draw path

API Layer

API GatewayHTTP API, Lambda proxy integration

→

API LambdaFastAPI + Mangum, python3.12

→

DynamoDBdraws, subscribers, saved picks, meta

AI Trend Analysis

Analytics Enginehot / cold / overdue frequency stats

→

Anthropic Claudeclaude-sonnet-4-5 blends trends into suggested sets

→

Backtest Enginevalidates every strategy against a random baseline

Scheduled Ingestion & Alerts

EventBridgecron matched to real Mega Millions / Powerball draw days

→

Ingest Lambdaseparate function, separate IAM role

→

DynamoDB + SES/Resendnew-draw detection triggers subscriber alerts

Infrastructure as Code

TerraformProvisions Lambda, API Gateway, DynamoDB, and EventBridge from HCL config

Two independent Lambda functions, each with its own least-privilege IAM role: the API Lambda never gets ingest permissions, and the ingest Lambda never gets exposed to the public API surface.

Rendering: React 19 SPA with a CloudFront Function handling SPA routing at the edge.

Compute: Two Lambda functions on python3.12, one for the public API, one for scheduled ingestion — never the same function.

AI: Claude blends hot/cold/overdue trend signals into suggested sets, explicitly disclaimed as entertainment only, never predictions.

Scheduling: EventBridge cron rules match actual draw calendars, not a generic polling interval.

Data: Four single-purpose DynamoDB tables — draws, subscribers, saved picks, and ingest metadata.

Security: The Anthropic key lives in SSM Parameter Store (SecureString), fetched at cold start — never in source, Terraform state, or a Lambda env var; each Lambda's IAM role is scoped to exactly what it touches.

Back to App Back to Portfolio