AI extraction accuracy - room-name matching
78.8%90.6%VERIFIED
18 fixtures · 773 ground-truth rows · production AI pipeline
AI ENGINEER · OSLO AND BRISBANE
I build AI that works in production. And I can show you the proof.
Most teams can ship a demo. The hard part is knowing whether the AI is actually reliable enough to put in front of a customer, and keeping it that way as models change. That is what I do.
AVAILABLE FOR NEW ROLES · JULY 2026
01 · WHAT I DO
What I do
Build production AI systems
GPT, Gemini, and Claude integrated into real production pipelines. Multimodal document and image extraction, agentic workflows. Not just prototypes.
Evaluate AI rigorously
Ground-truth datasets, regression evaluations, prompt and output testing, scenario-based behavioral testing, model telemetry and monitoring. Turns "I think it improved" into a measured number.
Build search and data systems
Elasticsearch architecture, relevance tuning, RAG and hybrid keyword/vector retrieval, indexing pipelines. At production scale, across large content bases.
Engineer full-stack
TypeScript, JavaScript, Python, Node.js, React, Next.js, REST APIs, CI/CD on GCP and Azure.
Own quality end to end
Automated test pipelines, API and E2E testing, security and performance testing. The practices that keep a system trustworthy as it changes.
02 · SELECTED EVIDENCE
Selected evidence
MEASURED RESULTS · SOURCED · NOT IMPRESSIONS
National library search platform
ZeroLive, 1M+ records searchableVERIFIED
Built from scratch · production in ~3 months · national institution
Amazon Oslo GenAI Hackathon
3rd placeVERIFIED
November 2025 · agentic AI solution · shipped in one day with a newly formed team
Test and release environment setup
Multiple developer-daysUnder one hourVERIFIED
Repeatable setup · automated pipeline · production engineering
03 · HOW I WORK
How I work
Treat AI output as something to evaluate, not to trust
Every model output is a claim to be checked against ground truth, not an answer to accept. That is my starting assumption, and it changes how my engineering tasks get done.
The ART of testing
Asserts - validate the right things. Readability - tests written in plain human language. Tested - cover the scenarios that actually matter. A methodology I have honed through production work.
AI-accelerated, engineering-disciplined
Use AI to build and validate faster, then hold the result to real engineering standards: tests, monitoring, evidence. Speed is not a reason I skip proof.
Build enough to prove the business case
Validate against real data and real users before over-polishing. Prototypes earn the right to become products. I do not over-engineer past the point where the value is clear.
Report plainly, especially the bad news
Results I share are understandable to non-technical stakeholders. Inconvenient findings arrive early and without hedging. The number is the number.
04 · ABOUT
About
I am an Australian software engineer. I spent fifteen years building quality into software for large-scale platforms, then applied that discipline to AI. I now build production AI systems and run the evaluations that prove they work.
I wrote Fantastic Elastic, a beginner-friendly book on Elasticsearch and Kibana.
05 · CONTACT
Get in touch
The easiest way to start a conversation is a direct email.