AI Engineer
Gervigreindarbrautin
Train the models everyone else just talks about.
What they'll learn
- The real ML workflow Take a model from raw data through training to honest evaluation, the way practitioners do it.
- Neural networks, hands on Build up an understanding from a single neuron to a working network — no hand-waving.
- Apps on large language models Use LLMs as a foundation and build real applications on top of them.
- Ethics and law that matter Cover the responsibilities, rules and legal limits a working AI engineer genuinely needs.
An ML project and an LLM application — each shipped with a professional model card.
The 13-week journey
-
Sprint 1 — Machine learning foundations
The real workflow on real datasets: features, train/test discipline, classification and regression, the metrics-honesty lab. Milestone: a defensible model on a dataset they chose.
-
Sprint 2 — Neural networks and language
From neuron to network, a small net trained and dissected; then language: embeddings, how LLMs are trained, retrieval patterns — grounded in the Icelandic language-tech story (Miðeind, Almannarómur). Milestone: a working LLM-powered prototype.
-
Sprint 3 — The build
Each student commits to an ML project or an LLM application and engineers it to standard: evaluation suite, bias probes, a model card; EU AI Act basics as professional context. Milestone: the project, evaluated.
-
Week 13 — Showcase
Demos plus a what-it-would-take-to-productionize talk — the question real engineers get asked.
What we cover
Every topic, unit by unit — so you know exactly what your child builds and learns.
Machine learning foundations
- The real ML workflow on real datasets
- Features, train/test discipline, leak-proofing
- Classification and regression with scikit-learn
- The metrics-honesty lab — where accuracy deceives
Neural networks & language
- From neuron to network: training and dissecting a small net
- Embeddings as meaning-geometry
- How LLMs are trained, plus retrieval patterns
- The Icelandic language-tech story (Miðeind, Almannarómur)
The build
- Engineering an ML project or LLM application to standard
- Building an evaluation suite and bias probes
- Writing a proper model card
- EU AI Act basics as professional context
Showcase & productionizing
- Error analysis and honest evaluation
- Demos with model/eval cards
- A 'what it would take to productionize' talk
A working machine-learning project and an LLM-powered application, each with a proper model/eval card.
The top of the school’s AI ladder and its proudest claim: teenagers doing real machine learning in Python — training models, evaluating them honestly, and building LLM-powered applications — framed by the most locally meaningful AI story there is: teaching machines Icelandic.
The hooks
Teen hook: “I train models and build with the same AI everyone talks about — most adults can’t do this.” Parent hook: “The scarcest skill line a CV can carry right now, taught hands-on.”
Who it’s for
Solid Python required — AI Makers, Machine Learning Lab, or placement. Comfort with math helps; the course teaches the intuition rigorously and the formalism honestly. Thrives: the analytical and the ambitious.
Outcomes — by the end, students can
Work data properly in numpy/pandas; train and compare scikit-learn models and choose metrics that don’t lie; explain neural networks from the neuron up and train a small one; build LLM applications (API orchestration, system prompts, embeddings, retrieval); evaluate models like a professional; place their work in context — the EU AI Act and Iceland’s language-tech effort.
Tools & compliance
Python notebooks, numpy/pandas/scikit-learn, a Keras-level intro for neural nets, school-managed LLM API access with caps, own accounts at 16+, public Icelandic and open datasets.
Where this course fits
The summit of the AI ladder; leads into the agent track, applied ML, and the Launchpad capstone.
Parent questions
Is this university-level material?
It overlaps with first-year university content, taught with more building and less lecture.
What math is required?
Framhaldsskóli math suffices; we build the intuition carefully and never hide behind notation.
Why does this matter for Iceland specifically?
Keeping Icelandic alive inside AI is a national project — students meet it as builders, not bystanders.
The first lesson is a free trial.
Book a no-commitment trial — pay nothing if it's not a fit.
Book a free trial