Ireland · Training, not prompting · Live online

Learn to train AI, not just prompt it

Prompting is a genuine skill, and it is a shallow one. It takes a weekend to get good at, the tools change every few months, and none of it tells you whether the answer you got is right. Training is the other half of AI, the half that decides what a model learns and how you would know if it had learned nothing useful. It involves choices no tool makes for you: what exactly are you predicting, what evidence would have been available at the moment of the prediction, how do you split the data so the test is honest, and what is the simplest possible approach your model has to beat? Ireland's 2024 National AI Strategy refresh makes a related point about people rather than models: "the skills needed for AI are not only digital skills, and the continued development of empathetic and cognitive skills will be crucial to the success of AI adoption in Ireland." This page shows the training half being done properly, on an Irish dataset anyone can download.

Book the free first class Five ways to fool yourself Ask on WhatsApp

Live teaching since 2020 · 10,000+ students · models trained, split and measured by the learner

In short

Training a model means deciding what to predict, choosing evidence you would genuinely have in advance, splitting the data so the test is fair, setting a baseline, and only then fitting anything. We teach that on real Irish data rather than on tidied teaching sets, because every important lesson lives in the mess: categories that repeat themselves as totals, missing values that are missing for a reason, and splits that let a model memorise instead of learn. Classes are live, in groups of five to ten at one level or one to one, for teenagers, students and adults. The first class is free, then USD 100 a month in a group or USD 150 a month one to one.

Start here

Three courses where you train the model

The mechanics of training, at three levels. Each card opens its syllabus.

AI and ML Masterclass course thumbnail

TRAIN / 01

AI and Machine Learning Masterclass

The full training loop for students and adults: features, splits, fitting, validation and the arithmetic of deciding whether a result is real.

Open the syllabus →
AI and Machine Learning for Teens course thumbnail

TRAIN / 02

AI and Machine Learning for Teens

The same ideas at secondary-school pace, with a first model the student trains, breaks on purpose and then fixes.

Open the syllabus →
Orange Data Mining Course for College Students course thumbnail

TRAIN / 03

Orange Data Mining

Training made visible, for learners who want to see the pipeline before they write one in code.

Open the syllabus →

Prompting or training

What training actually asks of you

Both skills matter. They are simply not the same work, and only one of them makes you able to judge a model.

Two different jobs
QuestionPromptingTraining
What do you control?The words you sendThe data, the labels, the features and the split
How do you know it worked?It looks rightIt beats a baseline you defined in advance
What can go wrong quietly?A confident wrong answerA score inflated by leakage or a bad split
What transfers when tools change?Some habitsAlmost everything

The first thing a learner meets in training is that the task has to be defined precisely enough to argue about. Predicting "how a place is doing" is not a task. Predicting one named number, for one month, from evidence that existed the month before, is.

The second is that you must be able to fail. A prompt never reports an error rate. A trained model does, and the number is only meaningful beside the simplest alternative anyone could have used instead.

The skills side of this is national policy. The 2024 refresh of Ireland's National AI Strategy notes that AI skills "are not only digital skills", and that cognitive skills will matter for adoption here.

The practical version of that sentence, for a learner, is judgement: knowing which questions the data can answer, and saying clearly when it cannot. The tool side is covered on learn to think, not just use AI tools.

Source: Ireland's National AI Strategy: AI, Here for Good, Refresh 2024, Department of Enterprise, Trade and Employment, read 16 September 2026.

A dataset to train on

Driving test pass rates: a training set every Irish family understands

The Central Statistics Office publishes driving test figures from the Road Safety Authority, monthly, by test centre and test category. It is open under a Creative Commons Attribution 4.0 licence, and it is a superb first training set precisely because it is awkward.

CSO table ROA30, Driving Tests Delivered and Pass Rate
AspectMeasured
ShapeAn exact cube: 4 statistics by 68 months by 18 test categories by 73 centres, which is 357,408 rows
PeriodJanuary 2021 to August 2026
The task we setPredict the car-test pass rate for a given centre in a given month
Usable rows for that task3,887, across 68 centres, once rows with no recorded pass rate are dropped
The number to beatPredicting the overall mean of 53.285% for every row gives an average error of 6.787 percentage points

Why a baseline first

That 6.787 is the whole point of the exercise. It is what you achieve by knowing nothing except the national average, and any model that cannot beat it has taught you only that your features are useless.

Students compute it themselves in the first session, before any model exists. It takes two lines and it reframes everything that follows.

And why this dataset

Because almost every household in Ireland has someone who has sat this test, so learners have real intuitions to check the model against, which is exactly when mistakes become visible.

We publish no ranking of test centres, and neither should a learner. The data supports a forecasting exercise; it does not support judgements about the places or the people in them.

Source: ROA30, Driving Tests Delivered and Pass Rate, published by the Central Statistics Office with the Road Safety Authority as source, listed on data.gov.ie as "Licensed under: Creative Commons Attribution 4.0", read and measured on 16 September 2026. Contains Irish Public Sector Data licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) licence. All counts above are ours, computed from the published file.

Five ways to fool yourself

Five traps in one file, each of them measured

These are not hypothetical warnings. Every number below came out of this file, and each one has ruined somebody's project at some point.

Measured traps in ROA30
TrapWhat we measuredWhat it costs you
A random splitA random 80/20 split put 67 centres in the test set, and all 67 also appeared in trainingThe model can memorise centres instead of learning; centre memorisation alone cuts the error by 24%
Time driftYearly mean pass rates ran 55.90, 52.19, 52.72, 51.59, 53.14 and 54.71 across the periodTraining on one era and testing across all of them is not forecasting
Missing for a reason81.48% of raw cells are blank, from 20.6% for the car test to 100% for one category; 43 centres never miss a value and 4 always doDropping blank rows silently deletes whole categories and venues
Totals hiding in the rows24,480 rows, 6.8% of the file, are totals, verified as exact sums of the detailed rows in all 68 monthsTraining on them means training repeatedly on the same observations
Columns that dieOnce you select the pass-rate statistic, three columns hold a single value eachA pipeline that encodes every column wastes features on constants

Split by group, not by row

Hold out whole test centres, or whole months, depending on the question. If your model will meet new places, test it on places it has never seen.

Ask why a value is missing

Here, small overflow venues are missing 55.7% of their values against 16.2% at permanent centres. The publisher does not say why cells are blank, so we do not claim a reason, but the pattern is clearly not random.

Read the names carefully

One centre name uses an en dash where three otherwise identical names use a hyphen. A split on the wrong character quietly mangles one row, and nothing warns you.

Working through these is the course. A learner who has met group leakage once, with a measured 24% gap attached to it, will check for it forever afterwards.

And when the honest model finally beats 6.787, they know exactly what the improvement is worth, which is the only thing that separates a result from a screenshot.

What it cannot tell you

The limits are part of the training

A model that predicts a number says nothing about why the number is what it is. Confusing those two is the most common mistake in applied AI.

Prediction is not explanation

A model can forecast a pass rate without any idea what causes it. Who sits the test where, when and after how much practice all sit outside this file.

Do not rank people or places

We do not publish league tables from data like this, and we teach students not to. The same restraint applies to staff, schools and patients in every other dataset they will meet.

Say what would change your mind

Every project ends with the conditions under which the model would stop working: a new category, a new venue, a change in how the figures are recorded.

The rungs

Five rungs of training a model

The order matters more than the algorithm. Most learners reach the last rung in three to five months.

The training loop, in order
RungWhat the learner does
1. Define the targetStates the exact quantity, the unit and the moment of prediction
2. Choose honest featuresKeeps only evidence that exists before the target is known
3. Split with intentHolds out groups or time periods, not random rows
4. Baseline, then fitMeasures the naive approach, then trains and compares
5. Test the resultChecks stability, looks for leakage, and writes the limits down

Where overfitting shows up

Usually between rungs three and four: the training score climbs while the held-out score stalls. Seeing it happen once, on your own data, teaches more than any definition.

The wider sequence of subjects is on the coding roadmap.

After this

Learners who want a finished, defensible project go on to build real AI projects, which works on a different Irish dataset.

Those heading for agents and language models take AI agents for students instead.

The catalogue

Nine live courses on training and evaluating models

By level. The free first class decides where you start.

I

Before training

The groundwork

TRAIN / BASE / 01

Python Masterclass

Enough Python to write a training pipeline yourself.

Open the syllabus

TRAIN / BASE / 02

Data Analysis Mastery

Reading a file properly before modelling it.

Open the syllabus

TRAIN / BASE / 03

Data Science for Teens

The same groundwork for secondary students.

Open the syllabus
II

Training

The loop itself

TRAIN / CORE / 01

AI and Machine Learning Masterclass

Features, splits, fitting, validation, judgement.

Open the syllabus

TRAIN / CORE / 02

AI and Machine Learning for Teens

A model a teenager trains and interrogates.

Open the syllabus

TRAIN / CORE / 03

Orange Data Mining

The pipeline seen before it is written.

Open the syllabus
III

Beyond one model

Scale and modern methods

TRAIN / NEXT / 01

Artificial Intelligence Masterclass

The field around the training loop.

Open the syllabus

TRAIN / NEXT / 02

Generative AI: LLMs, RAG and Agents

What training means for language models.

Open the syllabus

TRAIN / NEXT / 03

Data Science Masterclass

The analysis that surrounds any model.

Open the syllabus

How the classes run

Live sessions with the data open in front of you

Classes are live, with teachers working from India at a weekly hour set in Irish time. India does not change its clocks and Ireland does, so the gap is four and a half hours through the Irish summer and five and a half through the winter. Training sessions suit longer slots, so many learners choose a weekend time.

Weekday evening

Steady weekly progress, Irish time.

Weekend

Longer blocks for fitting and evaluation.

Quiet periods

Intensive weeks during holidays.

A free first class

You define a task and compute its baseline with a teacher, before any model is trained.

Five to ten per group

Learners at one level, comparing splits and scores on the same data.

One to one

For work data, a thesis, or a deadline that will not wait.

Baseline before model

No result is accepted in class without the naive score it has to beat.

Splits are checked

Every split is inspected for group and time leakage before training begins.

Limits written down

Each project ends with the conditions under which its model should not be trusted.

Student work

Models our learners trained

Four projects our students published, with their code and their evaluation. More are gathered on the student labs page.

NutriLife AI nutrition coach project screenshot

AI and ML

NutriLife

An AI nutrition coach that reads what you eat and works you toward a target.

by Bhavya · Open it

Misti AI chatbot for maths and coding screenshot

AI and ML

Misti

A chatbot that answers mathematics and programming questions, built and deployed by a student.

by Harshit · Open it

GuardianX AI internet safety assistant screenshot

AI and ML

GuardianX

An assistant that helps a young person recognise unsafe situations online.

by Vivaan · Open it

SkyCast weather forecast application screenshot

Web app

SkyCast

A weather forecasting site with live conditions for any location.

by Krish

Fees

Fees

One monthly fee in US dollars, identical in every country outside India, for any course on this page. Nothing to pay in advance for a term, and you stop whenever you choose.

Free first class

USD 0

no card required

  • A task defined and a baseline computed
  • An honest view of your starting level
  • No card taken
Book it

Group batch

USD 100

a month, billed in US dollars

  • Five to ten learners at one level
  • The same teacher throughout
  • Splits and scores reviewed weekly
  • A certificate at the end
Start here

One to one

USD 150

a month, billed in US dollars

  • A teacher to yourself
  • Your own dataset and question
  • Suited to work projects and theses
Enquire

What families say

Rated 4.9 across 547 Google reviews

Straight from Google: reviews written by our students and their families, never bought.

★★★★★

"The one step solution for my son. Modern Age Coders make learning coding so simple that kids love it. The teachers explain complex concepts clearly with practical exercises and interactive content."

Ria Mukherjee

Parent

★★★★★

"Modern Age Coders has been a game-changer for me. I struggled to grasp IT concepts and coding before joining, but their classes transformed everything. I can now confidently write complex programs with ease."

Samriddha Mondal

Student

★★★★★

"One of the most wonderful education centres out there. Education is not limited to school syllabus but focuses on skill development."

Vansh Agarwal

Student

★★★★★

"My child Dhairya is really enjoying the Modern Age Coders classes. This is his first online class and he eagerly looks forward to it. I can already see his improvement, and the teachers are very cooperative."

Sonam Oswal

Parent of Dhairya

★★★★★

"Modern Age Coders have wonderful teachers who teach in a clear, easy and practical way. The teacher boosts students' confidence and inspires them to learn without hesitation."

Sonu Goyal

Parent

★★★★★

"I highly recommend this computer coding class! The teachers are incredibly knowledgeable and passionate about coding."

Ritu Kedia

Parent

Questions

What learners in Ireland ask about training models

Is prompting not enough for most jobs?

For using a tool, often yes. For deciding whether to trust one, no. Training teaches you what a model can and cannot know, which is what judgement rests on.

Do I need a lot of maths?

Less than you fear. You need to be comfortable with averages, percentages and the idea of error. The heavier maths is taught only where a problem needs it.

What is a baseline, and why does it matter?

The simplest sensible prediction, such as always guessing the overall average. On the driving-test data that gives an average error of 6.787 percentage points, and any model has to beat it to be worth anything.

What is group leakage?

Letting the same group appear in training and testing. On this dataset, a random split put every one of the 67 test-fold centres into training too, so the model could memorise rather than generalise.

Where do I get data like this?

Ireland publishes plenty openly. This table is on the national portal under a Creative Commons Attribution 4.0 licence, which allows re-use with attribution.

Do I need a GPU?

No. Everything on this page trains in seconds on an ordinary laptop. Large models are a different subject and rarely where the learning is.

Can I train a model on my company's data?

Yes, in one-to-one lessons, and we will be careful about what leaves your machine. The method is identical; only the data changes.

What does it cost?

The first class is free. After that a group place is USD 100 a month and one-to-one lessons are USD 150 a month, in US dollars, with no enrolment fee.

When are the classes?

Evenings and weekends, at a weekly hour set in the free class. Our teachers are four and a half hours ahead of Irish time in summer and five and a half in winter.

What happens after I send the form?

We phone you at a reasonable Irish hour and arrange the free class. There is no charge unless you continue.

Elsewhere on this site

Related pages

Neighbouring subjects.

Build real AI projects in Ireland

A finished project, start to finish.

Data science course in Ireland

The data work underneath.

Learn to think, not just use AI tools

The judgement side.

AI agents course for professionals

Agents at work.

Best Python classes online in Ireland

The language for all of it.

AI and machine learning classes in Ireland

The national AI page.

Start here

Compute your first baseline in a free class

Leave a number and we will ring at a time that suits you in Ireland. In the free class you take a real Irish dataset, define a task with a teacher and work out the number any model would have to beat.

Prefer to read first? Try how we teach, the course syllabuses, and projects our students published.

WhatsApp us · +91 91233 66161 · contact@modernagecoders.com

WhatsApp is free from an Irish number and usually reaches us fastest. Ours is an Indian number, because our teachers are in India, and we keep no Irish office.

No card, no obligation, one call.

Keep exploring Modern Age Coders

Related courses and guides

Learn more

Free resources

From the blog

Start here