OCT 2026
📱 Let's see what you can do
iosswiftswiftuixcodebionic
Let's see what you can do. The key with AI is being able to articulate what you
want — the clearer the prompt, the better the outcome. I'd had a number of ideas for
iOS apps over the years but could never find the time to commit to creating them.
My first trial: I wrote a paragraph explaining one of the ideas, described the
features I wanted, and clicked send.
The process is very iterative and interactive, giving good feedback at every stage.
At the end of a few hours of collaborating, I had an app ready — one I'll try to
publish, and I'll update here once it's launched.
Stack: Xcode, GitHub, Bionic, Swift, SwiftUI.
Takeaway: The bottleneck for a shelf of app ideas was never the code — it was the commitment. A few hours of clear, articulated collaboration is the new "weekend project", and shipping beats polishing in your head.
OCT 2026
🍎 Latest silicon, latest models
mac-miniapple-siliconollamaqwen3.8-27bbionic
Latest silicon, latest models. I was lucky enough to purchase a Mac mini M6 after
reading up on the advancements in its AI capabilities — the ability to run a local
model with relatively low power consumption feels like a genuine game changer. My
previous GPU experiments ended up as a power-hungry, noisy setup: fine in a work
environment, but not something for the home.
Where to start? I asked myself that, and after lots of searches and a few rabbit
holes, I arrived at the stack I'm using now: Ollama, Qwen3.8 27B, Xcode, GitHub, and
Bionic.
Takeaway: The best local rig isn't the one with the most fans — it's the one quiet enough to keep at home. Efficiency flips the math: a model you'll actually run all the time beats a bigger one you can't bear to leave on.
AUG 2024
🖥️ Sourced a GPU — now the fun really begins
gpulangchaingpt-oss-safeguardagentic-aireact
Sourced a GPU, and now the fun really begins. One of the key challenges in my role is
keeping operations reliable and secure for our customers — and that work generates a
wealth of information and operational data. This is where local AI earns its keep:
pointed at that data, it unlocks the huge volume of historic logs and events
that were sitting there, unread, for years. And because the models run locally, none
of it leaves the building — for operational and customer data, that's what keeps it
genuinely secure.
Stack: LangChain for the orchestration, OpenAI's
gpt-oss-safeguard for the model, agentic AI for workflow automation, and React.js
on the front end.
Takeaway: Your operational history is a goldmine — but a goldmine you can't hand to a cloud API. Local models let you mine it without compromising the one thing operations can never trade: security.
OCT 2023
🎬 The video that lit the fuse
heygenvideo-translationlip-synclocalisation
My journey exploring what AI could actually do started when I came across a
company called HeyGen and their use of AI to translate video into a number of languages
while maintaining the lip sync. I work across multiple regions, and the ability to
create credible, localised content lit the fuse for everything since. At the time the
content had to be batched and was entirely cloud-based — and I started to think:
what's next? Real-time, local creation?
Here's an example of what I
made at the time.
Takeaway: Local AI journeys rarely start with a terminal. They start with a polished cloud demo that makes you ask what it would look like on your own machine — in real time, and private.