An Enterprise-Grade AI Agent Built on Amazon Bedrock AgentCore
A personal project modeling a real enterprise use case — building a production-grade customer support agent on Amazon Bedrock AgentCore, end to end.
I designed and built a customer support agent automating inquiries about product specs, return policies, warranty status, and troubleshooting, using Amazon Bedrock AgentCore (Runtime, Gateway, Memory, Knowledge Base) with the Strands Agents framework. Tools reach real data via DynamoDB and an S3 Vectors Knowledge Base, memory persists across sessions, Cedar policies govern access, and users authenticate via Cognito. The FastAPI backend and Next.js frontend are provisioned via Terraform.
- AgentCore Runtime Tools
- S3 Vectors Knowledge Base
- Persistent Conversation Memory
- Cedar-Based Authorization
- Dual Authentication Flow
- Next.js Chat Frontend
- Terraform IaC & CI/CD
Project Summary
This project demonstrates a full enterprise-grade agentic architecture on AWS — from tool orchestration and knowledge retrieval to fine-grained authorization and observability — packaged as a working customer support agent rather than a toy demo. It shows how to combine Amazon Bedrock AgentCore's Runtime, Gateway, and Memory with a real data layer (DynamoDB, S3 Vectors Knowledge Base) and enterprise-grade security controls (Cedar policies, least-privilege IAM, Cognito authentication), all deployed through a reproducible, Terraform-driven CI/CD pipeline with CloudWatch and X-Ray observability built in from day one.
