BC + AI

Draft preview. This page is up for review, not for booking. The program is agreed and the four dates are held, but registration is not open and the instructors are still confirming their bios and the detail of the curriculum.

BC + AI · Local and open models

Run real AI on hardware you already own.

Four weeks, online, in a room of twenty. You start with a model on your own machine and finish with a set of agents that read your mail, watch your files, and do the boring work overnight. None of it leaves your network.

This is the course for everyone who has been told the future of their work is a subscription to somebody else’s data centre, and would like to check that claim for themselves.

Tell me when registration opensSee the four dates

Registration for the first cohort opens September 30, 2026.

8 GB
is enough memory to start
4
weeks, 90 minutes a week
20
seats, so everyone gets unstuck
0
API keys required

The point of the whole thing

Where your work actually goes

Where your work goes with a local modelYour documents, mail, prompts and the model itself all sit inside your own machine. The only connection leaving your network is a one-time model download from the open model hub.Your machineYour documentsYour mailYour promptsAgents you builtThe model, running locallyonceOpen model hubdownload the weightsNot in the pictureNo API keyNo per-token billNo terms of service
One dashed line leaves your network, and it only runs once. After the download you can turn the network off entirely and everything on the left keeps working. That is the whole argument. The rest of this page is how you get there.

Who this is for

Three kinds of people this is built for

Your policy will not let you paste it into a chatbot

Teachers, public sector staff, clinicians, lawyers. You have been told not to put student names, case notes, or client files into a hosted assistant, and you concluded AI was therefore closed to you. It is not. It just has to run somewhere you control.

You would rather own a tool than rent one

If your workflow depends on a subscription, a rate limit, and a model that can be deprecated out from under you, this is the course that gets you off that footing. The model you download in week one still runs in five years.

You want to know how it actually works

Not prompt tips. What a model is, where it runs, what it costs in memory rather than tokens, and how to wire small ones together into something that quietly does real work while you are asleep.

The question everyone asks first

Can my machine actually do this?

Almost certainly yes, and by a wider margin than people expect. Four bit quantisation shrinks a model to roughly a third of its original size for very little loss, which is why a laptop bought in the last few years now runs something genuinely useful.

Pick your memory and see what fits. We do this properly in week one, on your actual hardware.

How much memory does your machine have?

A base MacBook Air, or most laptops bought in the last few years.

You can run

8Bparameters, four bit

~5 GB model8 GB available
Models that fit
Llama 3.2 3B, Qwen 2.5 7B, Gemma
What that is good for
Summarising, drafting, tidying notes, sorting and labelling mail, and the small always-on agents this course ends with.

Sizing assumes the four bit quantisation Ollama pulls by default, with headroom left for the rest of your desktop. If your machine is smaller than the smallest option here, you can still take the course. Lab time runs on ours.

The four weeks

Every week ends with something running

  1. 01

    Week 1

    Get a capable model running on your own machine

    Ollama and LM Studio, and why four bit quantisation is the reason a serious model now fits on an ordinary laptop. You size your own hardware, pull a model that fits it, and get it answering. We finish by turning the network off and watching it keep working.

    Ends with: A local model, running offline

  2. 02

    Week 2

    Make it yours

    Fine tuning a small open model on your own material, live, in about the time it takes to have a coffee. You give it a voice and then do the unglamorous part: check whether it actually got better at your work, or just different.

    Ends with: A small model tuned on your own material

  3. 03

    Week 3

    Agents that do the work you keep putting off

    Pointing a local model at the jobs you would rather not do: reading the inbox, watching a folder, drafting the summary, keeping the schedule. Small single purpose agents, running on a timer, none of it leaving your network.

    Ends with: Two or three agents running on a schedule

  4. 04

    Week 4

    Put it together, and know its limits

    Assembling week three into one orchestrator on open source you can read and change. Then the honest part that most AI courses skip: where local models are still worse, how to tell, and how to hand those jobs to a hosted model deliberately rather than by default.

    Ends with: Your own local AI orchestrator

You leave with

A machine that works for you, not a certificate you file away.

Every BC + AI program is built around something you can still use a month later. For this one that is an orchestrator running on your own hardware, on open source you can read and change, doing work you chose for it.

You also get a certificate from the association, which is genuinely useful for some employers and is not the point.

  • Size any machine in front of you and say what it can and cannot run.
  • Run a capable open model with the network off.
  • Fine tune a small model on your own material and tell whether it improved.
  • Build agents that read, summarise, and act without a hosted API in the loop.
  • Assemble them into one orchestrator that keeps running after the course ends.
  • Say honestly when a local model is the wrong tool, and reach for a hosted one on purpose.

Who is teaching it

Two people in the room, not one

Twenty people doing hands-on work will get stuck at different points in the same ninety minutes. So one person teaches and one works the lab, which is the difference between a class you follow along with and a class where your own machine ends up working.

Portrait to come

Dominic dos Santos

InstructorCampbell River, BC

Teaches all four sessions and owns the curriculum, from the first model you pull to the orchestrator you leave with.

Dominic brought this course to BC + AI and runs it end to end. He works with local models and agents on a multi-GPU workstation he calls the beast, which is the same machine your lab time runs on, so the hardware you are learning on is hardware he uses daily rather than a rented demo box.

  • Owns and runs the multi-GPU lab machine used for hands-on time
  • BC + AI member
  • Proposed and authored this curriculum
LinkedIn

Still to come from Dominic: Headshot, Bio in his own words

Portrait to come

Jeremy Bayley“Puma”

Lab leadCampbell River, BC

Second pair of hands during lab time, so twenty people can get stuck at once and nobody waits out the session.

Jeremy is one of the earliest members of BC + AI and a working video producer who builds his own local AI tools for film and game production, from a workflow orchestrator to a library for embedding local models. He also wrote the free book this course points you to afterwards. In this cohort he takes the questions that would otherwise stop the class.

  • Wrote Training Your Own Models on One 24 GB GPU, a free 40-chapter book
  • BC + AI founding member #11
  • Founder of VidBacon, a Vancouver video production company
BC + AI member profile

Still to come from Jeremy: Headshot, Bio in his own words

After week four

Training Your Own Models on One 24 GB GPU

The course gets local models running on your machine and walks you through training one small one. Jeremy’s free book goes much further into training. It starts with a three-parameter classifier and works up through language, vision, speech and sound, with 57 companion scripts. It is a handbook, not a benchmark, and it says so: its memory figures are worked calculations, not measured runs.

Read it free Source on GitHubExplore the book →

What it costs

CA$600

BC + AI members pay less, and so does anyone who registers early. Both reductions are applied at registration. Membership costs $340 CAD / year and carries weekly office hours, the community channels, and member rates on everything else we run.

When it runs

Four cohorts, booked a year ahead

If the first one does not suit you, take the next. We schedule a year out so you can plan around it, and we skip December and August because nobody wants to learn anything in those months.

  • Cohort 1

    First run

    November 11 to December 2, 2026

    Wednesdays Nov 11, Nov 18, Nov 25, Dec 2 · 90 minutes live on Zoom.

    Four weekly online pilot sessions.

  • Cohort 2

    Planned

    February 17 to March 10, 2027

    Wednesdays Feb 17, Feb 24, Mar 3, Mar 10 · 90 minutes live on Zoom.

    Four weekly online sessions.

  • Cohort 3

    Planned

    May 12 to June 2, 2027

    Wednesdays May 12, May 19, May 26, Jun 2 · 90 minutes live on Zoom.

    Four weekly online sessions.

  • Cohort 4

    Planned

    September 8 to September 29, 2027

    Wednesdays Sep 8, Sep 15, Sep 22, Sep 29 · 90 minutes live on Zoom.

    Four weekly online sessions.

Early registration for a cohort closes two weeks before it starts. For the first cohort that is Monday November 2, 2026.

How it runs

Online, but hands on real hardware

The awkward part of teaching local AI online is that not everyone has a big machine yet. So we give you one. Lab time runs over a VPN into a real multi-GPU workstation, and you can follow along on it before deciding what to buy for yourself.

Format
Four weekly sessions, online
Each session
90 minutes live, plus about 90 minutes of your own work between
Class size
20 people
Labs
Hands-on time on a real multi-GPU machine, reached over a VPN
Recordings
Every session is recorded and stays with the cohort
You need
A reasonably recent laptop or desktop. We will tell you what yours can carry before week one.

You are not doing this alone

The cohort ends. The room does not.

Everyone who finishes lands inside BC + AI: weekly office hours, monthly meetups across the province, Film Club, and a few hundred people working on the same problems. That is usually where the second thing you build comes from.

Who runs this

BC + AI Ecosystem Association

A room became an association. BC + AI is a nonprofit of 300+ members building British Columbia’s AI industry from the ground up, through monthly meetups across the province, weekly office hours, working groups, and certification programs like this one.

Courses are run with the practitioners who know the material. Instructors keep ownership of their curriculum and their teaching; the association handles registration, the platform, and the room you land in afterwards. What the association earns goes back into running the ecosystem.

Join the associationWhat membership includes

Members pay less for this course

CA$150off the seat

Membership is $340 CAD / year, so a single seat covers a good part of it before you have used anything else. Registering early reduces the price again, and the two stack.

Every other program
At least 25% off BC + AI-produced events, including the other three certifications.
Every week
Office hours, coworking, working groups, and the community channels.
Every month
Meetups around British Columbia, many of them included with membership.

Twenty seats, and the first one has not opened yet.

Registration for the November cohort opens September 30, 2026. Tell us now and you will hear before it goes out to the wider list, which matters when there are twenty seats and two discounts that stack.

Tell me when registration opensSee the other programs