// Sydney, Australia

Michael Zhang

Principal AI Engineer

I take LLM products from zero to production — architecture, guardrails, evals and the teams that ship them. At MYOB I built the company's first generative AI products, now live in market, and the AI engineering foundations the wider organisation builds on.

See my work LinkedIn ↗
michael@prod — zsh

$ whoami

principal-ai-engineer · 8+ yrs ML → GenAI

$ ls ~/shipped/

ai-assistant/ business-insights/ ai-bas/

$ make evals

safety · accuracy · relevance — PASS

$ ▊

0GenAI products, zero → one 0rounds of penetration testing to security sign-off 0+years building ML systems

01 /work

Selected work — LLM products in production

Three generative AI products built from first spike to launch, in a regulated financial context.

◐ Live in production

AI Assistant

Architect · Tech Lead · EM

MYOB's AI chat experience, embedded in the product. Two capability layers: general accounting knowledge, and agentic queries over customers' own financial data — ledgers, accounts and transactions in real time.

  • Built the first spike that proved the product; made the core architecture calls
  • Defence-in-depth guardrails; three rounds of penetration testing to sign-off
  • Evaluation and observability stack (Langfuse, DeepEval) with accuracy as the gate for financial answers
LangChainClaudeLangfuseAzure
∑ Live in production

Business Insights

Engineering Manager · Architecture

Reads ledger data and generates clear, human-readable commentary on business performance for SME owners — with guardrails for tone and financial consistency before any insight is accepted.

  • Two-stage architecture: daily metrics pipeline → on-request LLM generation
  • Kubernetes cron, SQS + DLQs, Postgres, Claude via AWS Bedrock
  • Leading the refactor onto the AI chat platform as a unified Insight Agent
BedrockKubernetesSQSPostgres
myob.com/ai-accounting ↗
⚙ First LLM product at MYOB

AI BAS

Lead ML Engineer

Agentic automation of Business Activity Statement preparation for Australian SMEs — MYOB's first LLM product, built from zero to one.

  • Built the entire GST classification backend with plain-English explanations
  • Python/FastAPI orchestration on Kubernetes, grounded in live ledger APIs
  • Introduced the org's foundational LLM tooling: Bedrock, Guardrails, Langfuse
FastAPILangfuseGuardrailsK8s
myob.com/ai-bas-software ↗

02 /experience

Experience

2025 — now

Principal AI Engineer · MYOB

Technical lead and engineering manager across MYOB's agentic AI initiatives. Built the first working versions of the company's GenAI products now live with SME customers, established the AI engineering foundations for the wider organisation, and hired and led the teams that carried them to launch — including DevAI, internal AI development environments used by engineers across the company.

2024 — 2025

Platform Lead · Endeavour Group

Org Admin for the GCP environment underpinning the entire data platform, enabling parallel builds for 10+ squads. Delivered $1.5M+ in cloud cost savings by redesigning BigQuery slot structures; led Databricks contract negotiations and platform security initiatives directly with Google.

2022 — 2023

Senior Machine Learning Engineer · Endeavour Group

Model APIs and ML pipelines powering personalised recommendations across email, web and mobile. Led the "Buy Given Click" model to production — a 5% purchase-rate uplift over 10 weeks of A/B testing.

2020 — 2022

Lead Machine Learning Engineer · Sapia.ai

Led the ML Engineering team behind Sapia's candidate assessment product — ranking, plagiarism detection and feature systems on AWS Lambda, Fargate and ECS. Introduced fairness testing for models.

2016 — 2019

Data Scientist & NLP Researcher · Hudson / University of Melbourne

NLP research on information extraction (MPhil, University of Melbourne) — publication extraction algorithms beating prior benchmarks by 10% — and data science for recruitment analytics.

03 /speaking

Speaking

AI Engineer Melbourne · Web Directions · June 2026

From Zero to Production: How 15 Engineers Shipped a Production LLM Product with AI Coding Tools

How a team of fewer than 15 engineers took an AI-powered chat experience from zero to production, embedded directly inside MYOB — covering evaluation, guardrails, latency and hallucination management, and how AI coding tools like Claude Code and Cursor acted as genuine force multipliers along the way.

Talk details ↗

04 /research + recognition

Research & recognition

/publications

/awards

  • 2026 Hello Tomorrow Recognition Award · MYOB
  • 2023 Superstar Spotlight Award, Advanced Analytics · Endeavour Group
  • 2021 Universal Achiever Award · Sapia.ai
  • 2019 Conference Travel Scholarship · Google
  • 2015–18 Melbourne International Research Scholarship

/certifications

  • GCP Professional Cloud Security Engineer
  • GCP Professional Cloud Architect
  • GCP Professional Machine Learning Engineer
  • AWS SysOps Administrator & Developer — Associate
  • DBX Databricks Lakehouse Fundamentals

/education

  • University of Melbourne MPhil, Computer Science
  • Tsinghua University BEng, Computer Science
  • UW–Madison Software Engineering exchange

05 /stack

Tools I ship with

genai
LLM agentsLangChainClaude / BedrockMCPLangfuseDeepEvalRAGGuardrails
lang
PythonTypeScriptNode.jsJavaC/C++
cloud
AWS Bedrock · Lambda · ECS · SQSGCP BigQuery · Vertex AI · Cloud RunAzure
infra
KubernetesDockerTerraformFastAPIGitHub ActionsOAuth 2.0
data
SnowflakeDatabricksPySparkAirflowPostgreSQLElasticsearch

06 /contact

Let's build something
that actually ships.

Open to conversations about AI engineering, LLM products in regulated industries, and speaking.

reveal email LinkedIn ↗