01Profile
Melbourne · Available
10+Years of engineering depth
Production AI, cloud platforms and dependable software systems.
View profile↘Senior AI · Full-stack · Cloud engineering
Engineering AI systems for the real world.
Explore systems
Production AI, durable automation and cloud-native software from architecture to operation.
02Capabilities
03Selected systems
06
Four confidential freelance systems plus two private research initiatives focused on trustworthy AI engineering.
AI, software, cloud and technical leadership across complex systems.
Sole delivery from discovery and architecture through production handover.
Agents, MCP, RAG, automation, evaluation, guardrails and human control.
Orders supported by event-driven systems designed for reliability and recovery.
01 / Selected systems
Applied engineering
Proof that advanced AI can still be accountable, observable and useful.
01
MCP Postgres Scout
A secure database-introspection agent built with Amazon Bedrock and AgentCore, exposing governed PostgreSQL tools without granting destructive access.
Open case study ↗
02
Infrastructure Monitoring Copilot
An AWS operations agent that correlates telemetry, investigates incidents and proposes evidence-backed remediation while engineers retain authority.
Open case study ↗
03
AI Claims Triage
A governed claims agent using AWS document intelligence, retrieval, policy and mandatory human review for sensitive decisions.
Open case study ↗
04
Omnichannel Order Orchestration
A deterministic order platform with an AgentCore operations layer for investigation, exception handling and natural-language operational insight.
Open case study ↗05
Personal research initiative · Public description intentionally abstracted
Agentic Delivery Control Plane
A private research program exploring how coding agents can work 24/7 at high speed while remaining bounded by repository contracts, security policy, independent review, deterministic CI gates, evaluations and human control.
Explore the research program ↗06
Personal research initiative · Public description intentionally abstracted
Enterprise Capability Fabric
A private research initiative exploring how organisations can create model-agnostic AI agents that answer from approved sources and execute governed actions through typed capabilities, durable workflows, policy and evidence.
Explore the research program ↗02 / Capabilities
One senior partner
From the first architecture decision to the system your team operates.
AI agents & intelligent products
Agent workflows, RAG, MCP, tool calling, memory, browser agents, evaluations, guardrails and human approvals.
LangGraph · OpenAI · Claude · Gemini · Bedrock · AgentCoreAutomation & durable execution
Long-running workflows, approvals, retries, idempotency, rate controls, replay, scheduling and operational recovery.
n8n · Temporal · Inngest · Trigger.dev · Step FunctionsFull-stack SaaS engineering
Modern customer products, internal platforms, APIs, real-time collaboration, authentication and operational interfaces.
TypeScript · Node.js · React · Next.js · NestJS · PostgreSQLCloud, integration & reliability
Event-driven architecture, cloud infrastructure, enterprise integration, observability, security and cost optimisation.
AWS · Cloudflare · GCP · Azure · Kafka · Terraform · CDK03 / Control plane
Production principles
Autonomy becomes valuable only when the boundaries are engineered.
CONTROL / 01
Every agent action needs identity, evidence, permissions, validation and a recovery path.
- Identity
Know which user, agent and workflow is acting.
01 - Policy
Restrict tools, data and environments by explicit rules.
02 - Evidence
Make outputs traceable to sources and tool results.
03 - Approval
Keep humans in control of sensitive or irreversible actions.
04 - Evaluation
Measure task quality, safety, latency and cost continuously.
05 - Recovery
Design retries, fallbacks, replay and safe failure behaviour.
06
04 / Field notes
Technical perspectives
Ideas for building AI systems that remain dependable after launch.
A2A v1.0: the enterprise boundary between agents
Why A2A and MCP belong in different layers, and what reliable cross-agent coordination actually requires.
Read field note ↗Loop engineering is control-system design
Moving from prompting an agent to designing the loop that discovers, delegates, verifies and remembers work.
Read field note ↗Graph engineering: the control plane above AI agent loops
Why prompts, loops and graphs are nested layers, and how production graphs make work, state, authority and recovery explicit.
Read field note ↗Context is the operating system of an agent
A production strategy for instructions, tools, state, compaction, memory, provenance and context budgets.
Read field note ↗RAG in 2026: choose the least complex system that wins
When basic, modular, graph and agentic retrieval are justified, and when more retrieval only adds cost.
Read field note ↗Agent memory is execution state, not a vector database
Why long-horizon agents need branch-aware state, validated summaries and deliberate forgetting.
Read field note ↗The harness around the model is the product
Planner, generator and evaluator patterns for long-running software agents that can continue safely.
Read field note ↗MCP at scale without context-window bankruptcy
Progressive tool discovery, code execution, sandboxing and data minimisation for large tool ecosystems.
Read field note ↗Evaluate agent behaviour, not just the final answer
A practical scorecard for task success, tool use, policy compliance, recovery, latency and cost.
Read field note ↗Fine-tuning in 2026: adapt behaviour, not changing facts
A decision framework for SFT, LoRA, preference optimisation and agentic RL, including when not to train at all.
Read field note ↗Open weights in production: choose the model, own the system
Licensing, benchmarking, quantisation, serving and operations for vLLM, SGLang and local model workflows.
Read field note ↗Architecting a private LLM platform for on-prem and air-gapped environments
A production reference architecture for model serving, policy, agent sandboxes, observability, evaluation and recovery.
Read field note ↗Amazon Bedrock AgentCore: the production agent platform map
How Runtime, Harness, Gateway, Identity, Policy, Memory, Evaluations and Observability fit together without confusing the model with the platform.
Read field note ↗Amazon Bedrock AgentCore vs Microsoft Foundry Agent Service
A service-by-service comparison of agent runtime, tools, identity, memory, evaluation, observability and enterprise governance.
Read field note ↗Amazon Bedrock AgentCore vs Gemini Enterprise Agent Platform
Where AWS and Google Cloud align, where their abstractions differ, and how to choose without reducing architecture to a logo checklist.
Read field note ↗Designing a portable agent platform across AWS, Azure and Google Cloud
A cloud-neutral control plane for agent identity, tools, policy, durable state, evaluations and evidence while still using each cloud well.
Read field note ↗05 / About
Engineering depth without losing sight of the business outcome.
I’m Gil John Ferolino, a senior AI and software engineering consultant, technical lead and hands-on engineer with more than 10 years of experience across retail, healthcare, government, insurance, ecommerce and enterprise consulting.
I connect the layers that are often treated separately: business requirements, architecture, application engineering, AI workflows, cloud infrastructure, security, observability and operational ownership.
The completed portfolio systems shown here are confidential freelance engagements I delivered independently. I owned the full lifecycle across discovery, architecture, product engineering, AI, cloud infrastructure, security, testing, deployment, observability, documentation and handover.
Based inMelbourne, Australia
EngagementsConsulting · Contract · Permanent
Ways of workingHands-on delivery · Technical leadership
Have a difficult system worth building?
Let’s turn it into dependable production.
Discuss a project or opportunityMessage me on LinkedIn
Engineering profile
GitHub · GJTeamOne