AWS MLSecOps Infrastructure

Companies ship AI fast but rarely secure these systems against attacks specific to language models. This project demonstrates a defense-in-depth architecture on AWS: if one layer is bypassed, the others hold. 0 misconfigurations across 44 deployed components.

Secured LLM API on AWS. Internet → AWS WAF (SQLi, XSS, IP reputation) → ALB → Go Proxy (rate limiting 25 req/min, multi-pattern prompt injection detection) → FastAPI (SHA-256 auth via Secrets Manager, PII masking, DynamoDB audit) → OpenAI API (private subnets). 44 Terraform resources · 0 Checkov failures · IAM least privilege · immutable ECR images.