Practical notes on AI, business and engineering.
What I've learned building real systems — written for the people who decide, not just the people who build.
Articles
Frameworks and questions every business faces when it starts working with AI.
- ArticleAI Strategy7 min read
Strategic AI Implementation Roadmap: From Curiosity to Operational Results
Most AI projects stall between the ChatGPT subscription and the P&L. A three-phase framework — Discover, Decide, Deploy — for Australian SMEs that want operational results rather than pilots.
- ArticleAI Strategy6 min read
Forward Deployed Engineering: what it is, and why a business should care
Frontier models made intelligence cheap; deployment is still hard. What forward deployed engineering is, how it works inside a business, and what to expect from an engineer who works this way.
- ArticleKnowledge systems2 min read
What a knowledge system could look like for a financial advice firm
An illustrative scenario: the knowledge-system approach applied to a practice that runs on client files, compliance records and advice history.
Build notes
How real systems were actually built — the decisions, the trade-offs, and what went wrong.
- Build noteImplementation4 min read
Build note: ingesting 139,000 documents nobody had organised
What it actually takes to turn a decade of client folders — emails, PDFs, photographed paperwork — into something a business can query. The pipeline, what broke, and why OCR was the expensive part.
- Build noteImplementation3 min read
Build note: the two-layer split that made an AI system safe to use
A regulated firm can't put client files into a chatbot. The architecture that let one use AI over its entire history anyway — and why permissions have to be a design decision, not a policy document.
- Build noteImplementation3 min read
Build note: why I put boring search under the AI, not a vector database
The retrieval layer for a firm-wide knowledge system: SQLite full-text search with domain synonyms, and an AI assistant with typed tools on top. Why the unfashionable choice was the right one.
One practical AI idea for your business.
No news roundup, no hype. Just one idea you can actually use.