Training course

Intermediate to Advanced Tailored to your team

Self-Hosted LLM Training

A practical course on running language models on your own hardware: what Ollama and its bigger siblings are for, what the machine actually needs to be, and where self-hosting stops being worth it.

Level: Intermediate to Advanced
Length: 1 or 2 days
Delivery: On site, Remote, Hybrid
Group size: 5 to 20 participants

See all courses

Booked privately for your team. Dates and length are agreed when you book.

What your team will be able to do

Every module ends with something participants can repeat on Monday morning, in your codebase and against your conventions.

Know when it pays

Judge honestly whether self-hosting solves your problem, or whether you are buying hardware to get around a contract clause.

Get a model running

Pull, run and prompt an open-weight model on your own machine inside the first morning.

Size the machine

Work out what memory, GPU and disk a given model needs before anybody orders anything.

Serve it to a team

Put a model behind an API that several applications can call at once without it falling over.

Keep the data inside

Show precisely where your prompts and documents travel, which is the whole reason most teams are in the room.

Compare the real cost

Put hardware, power and your own maintenance time next to an API bill and compare them properly.

Curriculum

Built around hands-on work rather than slides. The running order and the depth of each module are adjusted to the group before delivery.

70% hands-on

Deciding to self-host

From 1 day

The reasons that hold up, and the several that do not survive a spreadsheet
Open-weight model families, their licences, and what each is genuinely good at
Memory, GPU and what a model needs before it will even load
Quantisation: what you give up, and how much of it anyone notices

Getting one running

From 1 day

Ollama end to end: install, pull, run, and where it keeps everything
Choosing and swapping models without re-downloading half the internet
How local models behave differently, and how your prompts have to change
Calling it from your own code and from the tools you already run

Serving a whole team

From 2 days

Past one laptop: inference servers built for throughput rather than convenience
Compatible endpoints, so the code you already wrote does not care what is behind them
Batching and concurrency, and what happens when twelve people ask at once
Adding your own documents without sending them anywhere

Keeping it running

From 2 days

Who may call it, enforced rather than assumed
Model and server updates without a surprise change in the answers
Knowing it has got slow before the people using it tell you
What it costs per month once the initial enthusiasm has worn off

How long you book it for

Every length is drawn from the same modules. A longer booking is more of the course, not a different one.

Essentials

Intermediate

1 day

A day to get a model running on your own machine and to know which one it ought to be. The version for a team working out whether any of this is realistic.

Modules covered

Deciding to self-host
Getting one running

Core

Advanced

2 days

Adds serving a whole team rather than one laptop, and the operational work that follows it: access, updates, monitoring, and what it costs to keep alive.

Modules covered

Deciding to self-host
Getting one running
Serving a whole team
Keeping it running

Not sure which one fits? The intake call settles it, and the length can still change afterwards.

Built around your team

No two runs of this course are the same

The curriculum above is a starting point, not a syllabus. Before anything is booked I find out where your team actually stands, then rebuild the course around that. A team already running agents daily spends its two days somewhere completely different from one that installed the tool last week.

Step 1

Intake call

A call before anything is agreed: what your team already uses, what has gone wrong so far, and what you need them doing differently afterwards.
Step 2

Stack review

Language, framework, review process, CI. Exercises are rebuilt against your repository so nothing in the room is hypothetical.
Step 3

Curriculum reshaped

Modules get dropped, deepened or added. You see the adjusted outline and sign it off before the date is fixed.
Step 4

Delivery and follow-up

We run the course, then check in once the team has applied it to real work and the awkward questions have surfaced.

What gets adjusted

Depth of each module
Your stack and repository
Group size and seniority mix
Length and how days are split
English or German
On site, remote or hybrid

Is this course right for your team?

Who it is for

Platform and infrastructure engineers who will be asked to run this and want to know what they are agreeing to.

Backend developers integrating a model that is not allowed to leave the network.

Technically-minded data protection and security people checking whether self-hosting solves the problem they think it does.

What participants need beforehand

Comfortable on a Linux command line.
Basic familiarity with containers; nobody needs to be an expert in them.
A laptop with at least 16 GB of memory, or access to a machine that has more.

How the training is delivered

Same material either way. The format changes how much of it lands in the first week.

On site

I come to your office and work with the team in the room. Usually the most effective option for a first rollout, because the sideways questions get answered as they come up.

Remote

Delivered live over video in focused blocks rather than one long day, with shared screens and hands-on exercises on your own work.

Hybrid

A day on site to get everyone moving, followed by shorter remote sessions once the team is applying it to real work and hitting real friction.

Optional add-ons Certificate of participation

There is no public schedule. Every course is booked privately for a single team, and we agree the dates together.

Your trainer

Luka Breitig

Software engineer and AI coding trainer

I train engineering teams to work with AI coding tools, and I use the same tools to ship production software every week. The material comes out of building things, not out of a slide deck.

Before going freelance I spent six years as technical co-founder of a marketing agency, responsible for everything technical from automation infrastructure to coordinating a freelancer network. That is why the courses spend as much time on review, guardrails and rollout as they do on prompting.

Ships production software every week
Delivers in English and German
On site across Europe, or remote
Speaker at Nomad Summit, Chiang Mai

More about Luka

Questions about this course

The things teams ask before booking. If yours is not here, ask me directly.

No. This is an infrastructure course rather than a modelling one, and nobody trains anything during it. What it does assume is comfort with a Linux command line and with containers, because that is where the entire day is spent.

Not for the first day, and that is deliberate: small models run acceptably on a modern laptop, and the point worth making is that useful self-hosting starts smaller than most teams assume. For the serving material a GPU machine helps, and I can provide access to one if you have none.

The current open-weight families, chosen fresh before each delivery, because a fixed list would be wrong within two months. What does not change is the method for choosing one, and that is what the module is really about.

It solves one part of it and leaves other parts untouched, and being precise about which is a scheduled part of the day. Running a model locally means your prompts stay put. It does not answer who may query it, how long the logs live, or what happens to what it produces. Bring the question your legal team actually asked.

For some tasks plainly yes, for others plainly not, and the course is blunt about where the line runs. Classification, extraction and drafting against your own documents work well. Anything needing frontier reasoning still does not. You leave able to place your own use cases on that line yourself.

Five to twenty. Downloads and model loads take real minutes, so the day is paced around them, and a room larger than twenty turns the shared network into the thing everyone is waiting for.

Still not sure whether this is the right fit for your team?

Book this training for your team

Tell me the size of the group, the codebase they work in, and roughly when you want to run it. I will come back with a proposal.

Download trainer profile

Delivered in English or German, on site across Europe or remotely.

Self-Hosted LLM Training