Aws Agentic Ai Reference Architecture

AWS Agentic AI Reference Architecture

Production-grade reference architecture for resilient agentic AI workloads on AWS. The repository is designed as a portfolio-quality blueprint for Solution Architects building enterprise AI platforms with Amazon Bedrock, MCP tools, guardrails, identity, observability and Well-Architected practices.

Live portfolio / Portfolio ao vivo

• Production: AWS Agentic AI Reference Architecture
• Documentation: [Project docs](docs/architecture.md)
• GitHub: fernandofatech/aws-agentic-ai-reference-architecture
• Author: Fernando Francisco Azevedo · LinkedIn · GitHub

This public repository is part of a bilingual portfolio focused on solution architecture, AWS, AI, MCP/tooling, DevSecOps, and production-ready engineering practices.

Este repositório público faz parte de um portfólio bilíngue focado em arquitetura de soluções, AWS, IA, MCP/tools, DevSecOps e boas práticas de engenharia para produção.

Why this project exists

Most AI demos stop at a chat UI. Enterprise architects need more: resilience, security, cost control, operational visibility, governance and integration with real systems. This project documents and validates an AWS architecture for agentic AI that can survive production constraints.

Reference stack

• Amazon Bedrock for foundation models and managed AI capabilities.
• Amazon Bedrock Guardrails for responsible AI policy enforcement.
• MCP Tool Gateway for controlled tool access.
• AWS Lambda / ECS / Bedrock AgentCore-style runtime patterns for agents.
• Amazon OpenSearch Serverless or Bedrock Knowledge Bases for RAG.
• AWS IAM / identity broker for inbound and outbound authorization.
• Amazon CloudWatch, X-Ray and OpenTelemetry for traces, metrics and logs.
• EventBridge for asynchronous human-in-the-loop and audit events.

Architecture diagram

See [docs/diagrams/architecture.mmd](docs/diagrams/architecture.mmd).

Key architecture qualities

• Security-first: tool allow-listing, identity boundary, guardrails and audit events.
• Reliable: bounded retries, circuit breakers, fallback model strategy and human escalation.
• Observable: reasoning traces, tool invocation telemetry and quality evaluation signals.
• Cost-aware: token budgets, model routing and caching strategy.
• Enterprise-ready: ADRs, threat model, Well-Architected review and CI checks.

Run locally

python -m pip install -e . pytest
pytest -q

Portfolio positioning

This repository demonstrates Fernando Azevedo's focus on AWS, AI Engineering, DevSecOps, Well-Architected design and enterprise solution architecture for regulated environments.

Frontend

cd frontend
npm ci
npm run lint
npm run build

The frontend is a dependency-light static portfolio surface ready for Vercel deployment.

Operations

See [OPERATIONS.md](OPERATIONS.md) for GitFlow, Vercel secrets and security pipeline details.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fernandofatech
• Source: fernandofatech/aws-agentic-ai-reference-architecture
• License: MIT
• Homepage: https://agentic-ai.moretes.com

Install and usage instructions live in the source repository linked above.