Distributed systems &
AI agent platforms.
Secure, durable, built for real work.
I'm Steve Miles, a Perth-based software & hardware engineer. I build multi-tenant NATS platforms, zero-inbound-port remote operations tools, and durable multi-agent systems with explicit permissions and human approval gates.
Steve Miles
Companies I've worked with
#About
Two decades building connected systems that operate securely and recover cleanly.
My work spans embedded devices, telemetry, remote infrastructure, and multi-tenant cloud platforms. Recent systems combine distributed messaging, secure operations, and AI agents: NATS-native services, remote database and Docker tooling, and autonomous development workflows.
I'm most interested in the engineering behind trustworthy automation: tenant isolation, explicit permissions, at-least-once delivery, crash recovery, audit trails, and the context systems that let agents act usefully without giving them unlimited authority.
The model matters, but the harness matters more: durable state, typed gates, measurable outcomes, and human decisions at the points where judgment still belongs to people.
Hardware / IoT
- ESP32 / STM32 / RPi
- PCB design & prototyping
- Embedded firmware
- Sensor integration
Connected Systems
- NATS & JetStream
- Multi-tenant messaging
- Durable event pipelines
- Telemetry & remote operations
Software / Cloud
- JS, TypeScript & PHP platforms
- Containerised operations
- Cross-platform binaries
- SSR-first web applications
AI / Data
- Multi-agent orchestration
- MCP authorization
- RAG knowledge systems
- Context & evaluation harnesses
Products I'm building
#Featured work
Production systems built around security boundaries, durable state, and controlled automation. See all →
Levirge DevOps
Secure access to databases, Docker, logs, and services behind firewalls — one outbound connection, no VPN or inbound ports.
View platformLevirge Teams
Shared task management for humans and agents, with coordinated Lead, Coder, and Reviewer roles plus human approval gates.
View productLevirge Brain
Central, multi-domain knowledge that turns documents, code, and decisions into searchable, linked memory shared by every agent.
View product#Engineering principles
The constraints that make automation trustworthy in production.
Isolation is a boundary
Tenant isolation is enforced through accounts, subjects, ACLs, and dedicated integration tests — not inferred from naming conventions.
Delivery must be confirmed
Audit events are not considered written until the messaging system acknowledges them.
Agents need constrained authority
Useful agents need tools and context. Trustworthy agents also need bounded writes, explicit permissions, and human gates.
Recovery is a feature
Long-running workflows persist their state so a process crash does not mean starting the work — or the reasoning — again.
#From the blog
Notes from the bench and the terminal. All articles →
Right tool for the job: why my chat engine lost the reranking race
The same model ran 9× faster after switching engines. llama.cpp SYCL won the chat workload and lost reranking, on the same Intel GPU.
2026-07-26
The models were not the only thing under evaluation. Our harness was wrong too.
What broken gates, state restoration, tiny samples, and interrupted runs taught us about evaluating LLM agents.
2026-07-15
One knowledge base, two AI jobs, and the smaller local model won one
Why the best model for answering questions was not the best model for building the knowledge those answers depend on.
2026-07-14Have a project in mind?
Whether it's a fleet of sensors, a web platform, or an AI workflow — I'd love to hear about it.
Contact me