Payment Intelligence Agent

Payment Intelligence Agent (PIA)

Cloud-native, event-driven payment transaction intelligence agent. Ingests simulated payment events, detects anomalies via rules + LLM, and produces automated risk reports through Claude with autonomous tool calling.

[](./LICENSE)       

---

Table of contents

• Why this project exists
• High-level architecture
• Repository layout
• Prerequisites
• Quickstart
• License

---

Why this project exists

PIA is a **portfolio-grade** implementation of a production-ready autonomous analysis agent for payment transactions. It demonstrates:

• Hexagonal architecture with strict separation between domain, application and infrastructure.
• Event-driven ingestion using SQS with DLQ, idempotent consumers, and the Outbox pattern.
• Autonomous LLM agent — the Claude API is wired with a proper tool-calling loop (not a one-shot prompt), giving the model first-class access to domain-aware tools.
• AWS cloud-native deployment on ECS Fargate with IaC via Terraform.
• PCI-DSS mindset: zero PAN, tokenized card references, PAN masking in logs, SSM secrets, IAM least-privilege.
• Full observability: structured JSON logs, Micrometer/CloudWatch metrics, OpenTelemetry traces.

Architecture

C4 — C