Gil Ferolino Forward Deployed AI Engineer
Menu

Portfolio map

Explore systems, engineering capability and current AI thinking.

A production-focused view of the systems I build, the controls behind them and the research shaping what comes next.

Melbourne, Australia · Available for selected engagements Start a conversation ↗
← Selected systems

Enterprise integration · Amazon Bedrock · AgentCore · Event-driven AWS

Omnichannel Order Orchestration

An event-driven AWS order platform with Amazon Bedrock and AgentCore for natural-language investigation, exception handling and operational decision support.

Architecture: Amazon Bedrock + Amazon Bedrock AgentCore

EngagementConfidential freelance project under NDA
My roleSole End-to-End Solution Architect, Full-Stack and Cloud Engineer
TechnologyAmazon Bedrock · AgentCore Runtime · Gateway · ECS Fargate · EventBridge · Kafka · SQS
Proof signalDeterministic transaction processing with an accountable AI operations layer
Omnichannel order orchestration dashboard showing order flow and exceptions

The challenge

Orders arrive from multiple channels and must be validated, deduplicated, routed and fulfilled despite downstream failures, duplicate events, traffic spikes and partner variability.

My end-to-end ownership

I independently designed and delivered the platform from requirements through operational handover. I owned the product workflow, solution architecture, APIs, services, event contracts, Kafka processing, validation, deduplication, fulfilment routing, retries, recovery, operational interfaces, infrastructure as code, security, testing, observability and deployment.

  • Discovery and requirements
  • Product and solution design
  • Architecture and engineering
  • AI and data workflows
  • Cloud, security and CI/CD
  • Testing, observability and handover

System design

Event-driven services coordinate validation, inventory routing, dispatch and status updates using TypeScript, Kafka and AWS. Operational views expose throughput, exceptions, service levels and replayable recovery paths.

AWS order and agent architecture

The AI layer does not own the order ledger. Deterministic services remain authoritative while AgentCore gives operators a governed way to investigate and act on exceptions.

01 / ChannelsAPI Gateway + Cognito

Web, mobile, marketplace, POS and partner requests enter through authenticated, versioned contracts.

02 / Transaction coreECS Fargate services

Validation, deduplication, routing and fulfilment execute as deterministic services with idempotent APIs.

03 / Event backboneAmazon MSK or Kafka + EventBridge

Domain events fan out to inventory, fulfilment, analytics and partner integrations without coupling channels.

04 / RecoverySQS + dead-letter queues + Step Functions

Retries, delayed actions, replay, compensation and approval waits remain durable and inspectable.

05 / Agent layerAmazon Bedrock + AgentCore Runtime

The operations agent explains order state, correlates events and prepares bounded exception-resolution plans.

06 / Governed toolsAgentCore Gateway + Identity + Policy

Provides read tools by default and approval-gated mutation tools with scoped credentials and complete audit.

DynamoDB or Aurora stores authoritative operational state, while CloudWatch, X-Ray, AgentCore Observability and Evaluations measure service health and agent behaviour independently.

Production controls

Idempotency keys, bounded retries, dead-letter queues, replay tooling, observability, autoscaling and infrastructure as code support systems processing more than 50,000 orders per day.