Architecture Overview · Enterprise AI Delivery

Spec-to-Production Healthcare AI System on AWS

Full production AWS deployment generated entirely by AI from a specification — 9 CloudFormation stacks, 8 Lambda functions, 2 MCP servers, and 5 dashboards. Zero IaC lines written manually.

0 IaC Written ManuallyAWS Bedrock · ClaudeHIPAA Architecture

Click any component to explore it in the full case study.

Input
Product Specification (Spec-to-Prod)
Complete system specification: 13 infrastructure modules, 8 Lambda functions, 2 MCP servers, 5 dashboards, HIPAA compliance requirements, multi-agent orchestration design. Zero manual IaC written.
0 IaC lines written manually
AI-Generated Infrastructure Pipeline — 5 phases
Phase 1
Ingestion
S3 event trigger → file validator Lambda
›
Phase 2
Normalization
Step Functions 12-state ETL pipeline
›
Phase 3
PHI Masking
Macie → Comprehend Medical → deterministic masking
›
Phase 4
Graph + Vector
Neptune Gremlin + OpenSearch Serverless indexing
›
Phase 5
Agent Analysis
Bedrock orchestrator + 4 specialists + synthesizer agent
🔐
HIPAA Security Architecture
HIPAA tagging on all resources. VPC Flow Logs to CloudWatch (90-day retention). Least-privilege IAM roles per function. Structured logging. SNS alert thresholds.
HIPAA-tagged resourcesVPC Flow Logs (90d)Least-privilege IAMCloudWatch structured logsSNS alert rulesAudit trail per agent call
Output A
AWS Infrastructure — IaC
9 CloudFormation nested stacks. 8 Lambda functions. 2 MCP servers on ECS Fargate. All generated from spec by Claude/Kiro — zero manual IaC writing.
9 CFN stacks · 0 manual IaC
Output B
5 Operational Dashboards
6R Output · Data Quality · Custom Weights · ROI Engine · HITL Review. Live HTML dashboards generated alongside IaC, fed from agent analysis results.
5 HTML dashboards
9
CFN nested stacks
5
Bedrock agents
8
Lambda functions
0
IaC lines manual
StackAWS CloudFormationLambda Python 3.11AWS BedrockClaude SonnetNeptune GremlinOpenSearch ServerlessECS FargateStep FunctionsMacieComprehend Medical