01
Ready to try · 6 sections
Train your first AI model
Start with I²R. Make examples, train a tree and find where it goes wrong.
Keep: A first regression tree and a reason to keep the equation.
Open lesson →The learning path
Small steps. A model you understand.
Start without code. Build your first model, explore its limits and choose a useful engineering question.
Block A, lessons 1 to 4, is available now. Blocks B to F are coming soon while I review them. Work through the available lessons in order.
Your learning path
The lessons build on each other, from your first model to a tool you can use again.
Your account →The newsletter is optional and has its own signup.
Loading your saved progress…
Complete each lesson to unlock the next. Your account status is being checked.
What you can put into practice
Follow your progress from a guided example to a result you can use again.
- First model
- Frame a useful question and compare a first model with an equation.Still to practise
- Consistent data
- Check units, keep designs together and compare against a reference.Coming soon
- Python in practice
- Edit and rerun Python, export a table and plot, and adapt a model to another CSV.Coming soon
- Simulation to dataset
- Inspect documented simulation runs and calculate signal features.Coming soon
- Defensible evaluation
- Compare models, inspect mistakes and recognise unsupported predictions.Coming soon
- Reproducible tool
- Freeze a candidate, review reserved cases and package a reusable teaching tool.Coming soon
These milestones record completed course practice. A different case or another person reproducing your work provides further evidence; progress alone does not measure independent mastery or certify a model.
AUnderstand what a model does01 to 04 · 0 of 4 complete
Make a first model and choose a useful engineering question.
02
Checking your progress · 5 sections
Better examples, better questions
Find a gap in your examples, spend a small data budget and check what actually changes.
Keep: Save a coverage map and compare the original, reduced and added-example models on the same validation currents.
Review the public preview →Your account
03
Checking your progress · 5 sections
Check a model on a new design
Plan how to check a model on a new component design, then see why a different split answers a different question.
Keep: Save a complete design partition map and explain which evaluation supports the intended use, while keeping reserved designs out of both comparisons.
Review the public preview →Your account
04
Checking your progress · 6 sections
Choose your AI task
Turn an engineering idea into a brief that says what answer you need, how you will check it and when to request more work.
Keep: Download a practical brief with a decision, available inputs, simple reference, success criterion and limits, using the conduction-loss example to define the target.
Review the public preview →Your account
Block A mini-project · 10 guided steps
Loss Lab — Your first converter loss estimator
Bring the four lessons together: upload a converter simulation table, train and compare two estimators, check a complete new design and hand over an honest scope.
Keep: a working estimator, its prediction files and a use card that explains where it can—and cannot—be used.
Preview the mini-project →Checking your progress
Rate Block A or leave feedback
BPrepare data and compare05 to 08 · Coming soon
Repair a dataset and check an estimator against a simple reference.
05
Coming soon · 6 sections
Read your engineering data
Read a real teaching CSV, follow one simulated run and record what every column means.
Keep: Download a data dictionary with column roles, units, row meaning and the source of the reference values.
Coming soon06
Coming soon · 6 sections
Repair a dataset
Repair a deliberately damaged CSV using its original run records, keeping a reason for each change.
Keep: Download five repaired teaching rows and a cleaning log that preserves their source and meaning.
Coming soon07
Coming soon · 6 sections
Build a useful first estimator
Train two estimators, compare them with a constant and an engineering equation, and inspect where they fail.
Keep: Save a four-method comparison with errors in watts, physical checks and a clearly stated simulation scope.
Coming soon08
Coming soon · 6 sections
Improve without fooling yourself
Use validation for experiments, then freeze a choice before opening its final check.
Keep: Save a frozen model choice, a declared tolerance and a clearly labelled evaluation report.
Coming soonCLearn the Python you need09 to 12 · Coming soon
Run, change and download your own working notebook.
09
Coming soon · 6 sections
Run your first notebook
Run a calculation, change it, understand an error and reproduce your work.
Keep: Keep an editable notebook that calculates resistor losses from a clean start.
Coming soon10
Coming soon · 6 sections
Python for engineering calculations
Build a small calculation that accepts values, returns watts and refuses invalid inputs.
Keep: Write and test a reusable loss function with explicit units and a declared range.
Coming soon11
Coming soon · 6 sections
Turn a CSV into a plot
Read a small table, select comparable rows, convert units and draw your own graph.
Keep: Keep a notebook, a plot generated by Python and a transformed CSV with clear units.
Coming soon12
Coming soon · 6 sections
Train a model you can change
Connect a small training pipeline to a CSV, then adapt the inputs and retrain.
Keep: Keep a notebook that trains on grouped teaching data and exports predictions for another CSV.
Coming soonDWork with simulations and signals13 to 16 · Coming soon
Turn documented simulation runs into useful inputs and predictions.
13
Coming soon · 6 sections
From simulation to dataset
Trace a training row back to the converter settings, simulation run and extraction window that produced it.
Keep: Save a dataset manifest with accepted runs, exclusions and a precise reference definition.
Coming soon14
Coming soon · 6 sections
Cover the operating range
Choose valid operating conditions and discover gaps that simple minimum and maximum checks can miss.
Keep: Save a sampling plan with a data budget, reserved groups and a written domain of use.
Coming soon15
Coming soon · 6 sections
Turn waveforms into features
Select a meaningful time window and calculate waveform summaries that you can check by hand.
Keep: Save a reproducible feature table with units, window boundaries and execution groups.
Coming soon16
Coming soon · 6 sections
Build a simulation surrogate
Approximate a declared simulation output and compare the result with a physical reference inside its domain.
Keep: Save a conduction-loss estimator, its comparison and the conditions that need simulation review.
Coming soonEMake better decisions17 to 20 · Coming soon
Compare models, inspect mistakes and recognise unsupported predictions.
17
Coming soon · 6 sections
Classify simulated behaviour
Build a binary review-priority classifier using a declared ripple rule and inspect the mistakes it makes.
Keep: Save a logistic_regression classifier, its rule comparison and examples of misclassified runs.
Coming soon18
Coming soon · 6 sections
Choose which mistakes matter
Move a review threshold on validation and see how missed cases and unnecessary flags change.
Keep: Save a threshold choice with its assumptions, observed counts and reserved-evaluation plan.
Coming soon19
Coming soon · 6 sections
Compare stronger models
Compare a small number of ensemble models fairly, then judge whether their extra complexity is worthwhile.
Keep: Save a reproducible comparison covering grouped validation, errors, stability and observed runtime.
Coming soon20
Coming soon · 6 sections
Know when to stop trusting a prediction
Turn known model limits into explicit rules for accepting an input, requesting review or declining a prediction.
Keep: Save use and abstention rules with cases that demonstrate how the rules behave.
Coming soonFBuild a tool you can use again21 to 24 · Coming soon
Choose data, freeze a model and deliver a reusable tool.
21
Coming soon · 6 sections
Choose the next simulations
Spend a limited simulation budget on a development question, then check whether the new information helped.
Keep: Save a candidate-request plan and a fair before-and-after validation comparison.
Coming soon22
Coming soon · 6 sections
Build your own model
Adapt a real modelling decision and prepare a complete candidate for a later independent check.
Keep: Save your problem brief, data contract, comparison and frozen pipeline with a declared acceptance rule.
Coming soon23
Coming soon · 6 sections
Use your model again
Save a complete model, start clean and use it on a new CSV with clear handling of invalid requests.
Keep: Download a reusable package containing the model, code, example inputs and instructions.
Coming soon24
Coming soon · 6 sections
Engineering challenge
Evaluate a frozen tool on new cases and decide whether its evidence supports the proposed engineering use.
Keep: Deliver a reproducible tool and a use card with a defensible acceptance or insufficiency decision.
Coming soonKeep the engineering in view
Know where each result comes from.
The first exercises use equations. Later lessons use a documented teaching simulation of a buck converter. Neither is a hardware measurement. A simple physical calculation may still be your best tool.