Own your AI: Make AI-powered products built on custom and small LLMs — with Moiz

Most teams building something with AI start the same way: pick a big-name model, plug in the API, and hope the bill stays reasonable.
Moiz Penkar reckons that's often the wrong move — and he's bringing a laptop to show you why.

About the talk.

The default right now is to reach for an enormous cloud-hosted model. For a lot of real work, that's overkill.
A smaller model, fine-tuned on your own data, can be cheaper to run, easier to keep private, and considerably kinder to the planet. It also leaves your team genuinely understanding the thing they've built, rather than renting someone else's and hoping the terms don't change.
Moiz will walk through where each approach actually wins — this isn't a sales pitch, and frontier models have their place — and then get to the good bit: customising a model and running it live on ordinary consumer hardware. No data centre required.

Is this for you?
Probably, if any of this sounds familiar:

• Your team is trying to build AI into how you work, or into what you sell, and nobody's quite sure what you'd be committing to
• Your data can't leave the country — or the building — and that's already ruled options out
• You've looked at the API costs and had a quiet moment
• You want to run a model yourself and haven't known where to start
• You keep hearing "sustainable AI" and want to know if it means anything

You don't need a machine learning background. Moiz has condensed this from a longer session, and a decent chunk of it isn't about code at all — it's about how you decide. Engineers, team leads, product people and the plain curious will all find something here.

About Moiz

Moiz Penkar comes at this from HCI (Human Computer Interaction) research, and spends a good deal of his time making the case for a more mindful approach to new technology.
By day he's a lead front-end engineer, with consulting and contracting work behind him. He's spoken at UX Wellington a couple of times, runs a monthly series of TED-style talks and discussions, and has organised two TEDx events. He recently presented on modern front-end tooling at CodeCamp Wellington — and won the Claude Impact Lab hackathon, against a field that included a fair few AI specialists and emergency management experts.

On the night
5:15 PM — Doors open. Grab a drink and something to eat, say hello
5:30 PM — Moiz is up, for about an hour
6:30 PM — Q&A, around 20 minutes
6:50 PM — Nibbles, drinks, and the conversations that are usually the best part
7:30 PM — Wrap up

We're hoping to record the talk and Q&A for people who can't get to Wellington on the night. The camera will be pointed at the front, so if you'd rather stay out of frame, a seat toward the back does the job — or just have a word with us.

Thanks
A big thank you to Creative HQ for hosting us and providing the space for the evening. We're very grateful