/ James Baxter — Software Architect & Principal Engineer — Brisbane, Australia

I help engineering teams make AI coding agents dependable in real codebases — so they can deliver faster without losing architectural control.

I design the specifications, context architecture and delivery guardrails that turn AI-assisted coding from an experiment into a repeatable engineering capability. Behind that sits 20+ years of platform architecture, backend engineering and production diagnostics.

Discuss a problem Read how I work ↗

/ Services

/ 01

AI delivery systems

I design the specification, documentation and context architecture that allows coding agents to work reliably across real repositories — repository instructions, scoped context loading, documentation gates, acceptance criteria, and workflows that prevent generated knowledge from accumulating unchecked. In one bounded greenfield platform build, I measured roughly 12x my usual solo delivery rate using a specification-driven AI workflow; architecture, review and acceptance criteria remained explicit throughout.

» In public: VendKit — a Go tool that distributes agent instructions and other curated files across repositories, held to upstream by a checksum manifest and upgraded through review-gated pull requests.

12×
measured solo delivery rate — one bounded greenfield platform build, specification-driven AI workflow

/ 02

Platform architecture & modernisation

Distributed JVM systems (Java/Scala) and multi-tenant SaaS platforms on Azure and Kubernetes — declarative provisioning, GitOps reconciliation and per-tenant isolation, designed as platforms rather than one-off systems. I reshape systems at their architectural limits so they can grow again without a rewrite-the-world bet, and I treat developer experience as an architectural concern: one-command local stacks, ephemeral per-PR environments, deterministic builds.

» Designed a control plane where one operator action provisions a complete customer environment — Kubernetes cluster, managed databases, networking, identity and observability — reconciled continuously from Git.

/ 03

Production investigation

Failures that resist conventional debugging, traced to the JVM, the dependency graph or the packet.

» Resolved intermittent gateway 504s with no corresponding application traces by isolating corruption in a TCP timestamp option on the network path.

» Diagnosed a recurring production outage at a major bank by tracing connection-pool collapse to UDP lookups silently dropped by a firewall.

/ Ways to engage

01

Diagnostic sprint

A focused one- to two-week investigation producing findings, priorities and a practical implementation plan.

02

Architecture engagement

Fixed-scope design and validation for an AI delivery system, platform initiative or modernisation programme.

03

Fractional technical leadership

Ongoing principal-level architecture and hands-on engineering for teams that need senior capability without a permanent hire.

04

Production intervention

Intensive investigation of complex or business-critical failures, subject to availability.

» Most engagements begin with a paid discovery or diagnostic phase.

/ Track record

20+ yrs

of enterprise product engineering.

1 click

architected a multi-tenant SaaS platform where a single operator action provisions a complete customer environment across two-dozen-plus cloud resource types.

doubled the throughput of an enterprise product.

1 → 10

grew an engineering function from one developer to a ten-person cross-functional team.

/ Contact

If it’s ambiguous, high-stakes, or nobody can explain it, it’s probably my kind of problem.

james@baxter.consulting