Ransford's Notes

Core path

Your path through this level

Progress saves in this browser and syncs after you sign in

Completed

0 of 5

You will be able to

Explain the training loop in plain language and diagnose common failure modes like overfitting and leakage.
Design a basic evaluation plan (metrics + thresholds) that matches a real-world decision and risk profile.
Describe how data quality, features, and representation affect model behaviour in production.

Pick up where you left off

Training loop shape Features control behaviour Evaluation is evidence Deployment pattern Governance is enforcement

Optional

Full module map

Use this if you want the shape of the level before you start

Show

Module 2.1Training loop shape

Training is a loop: predict, measure error, update parameters, repeat with discipline.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain training loop shape in your own words and apply it to a realistic scenario.
Training is a loop: predict, measure error, update parameters, repeat with discipline.
Check the assumption "Loss matches the decision" and explain what changes if it is false.
Check the assumption "Validation is separate" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Overfitting. The model learns the training set, not the task. It looks strong and fails on new data.
Leakage. Information from the future sneaks into inputs. The score rises and trust collapses.

Module 2.2Features control behaviour

Representation choices decide what the model can notice and what it will ignore.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain features control behaviour in your own words and apply it to a realistic scenario.
Representation choices decide what the model can notice and what it will ignore.
Check the assumption "Meaning is preserved" and explain what changes if it is false.
Check the assumption "Bias is examined" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Proxy features. A feature can stand in for a sensitive attribute even if you never included it directly.
Schema drift. A feature changes meaning after a release. The model keeps working and starts guessing.

Module 2.3Evaluation is evidence

Evaluation is not one metric. It is evidence that the system is safe enough for the decision.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain evaluation is evidence in your own words and apply it to a realistic scenario.
Evaluation is not one metric. It is evidence that the system is safe enough for the decision.
Check the assumption "Metrics match harms" and explain what changes if it is false.
Check the assumption "You inspect failures" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Good average, bad edge cases. The average score can hide failures that matter most.
Threshold optimism. Thresholds set too aggressively create a system that looks good on paper and harms users.

Module 2.4Deployment pattern

Deployment is choosing a pattern that matches latency, cost, and safety.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain deployment pattern in your own words and apply it to a realistic scenario.
Deployment is choosing a pattern that matches latency, cost, and safety.
Check the assumption "Latency budget is known" and explain what changes if it is false.
Check the assumption "Costs are measured" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Hidden compute cost. Costs creep through retries, larger prompts, and wider retrieval. Measure and cap.
Missing fallbacks. When services fail, users still need a next step. Design for partial failure.

Module 2.5Governance is enforcement

Governance works when it is enforced by the system and measured by evidence.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain governance is enforcement in your own words and apply it to a realistic scenario.
Governance works when it is enforced by the system and measured by evidence.
Check the assumption "Owners exist" and explain what changes if it is false.
Check the assumption "Evidence is reviewable" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Policy only in text. A policy that is not enforced is a wish.
No change control. If changes are not tracked, you cannot explain failures or defend outcomes.

Optional

Planning and evidence

Objectives, timing, and CPD tracking

Show

If you want to start learning now, leave this closed. Come back when you want to plan your practice or keep evidence for CPD. This is guidance and it is not endorsed by awarding bodies. Standards mapping lives on the course overview page.

Learning objectives

What you will be able to do

1. Explain the training loop in plain language and diagnose common failure modes like overfitting and leakage.
2. Design a basic evaluation plan (metrics + thresholds) that matches a real-world decision and risk profile.
3. Describe how data quality, features, and representation affect model behaviour in production.
4. Select an appropriate deployment pattern (batch, real-time, retrieval-augmented) and explain the trade-offs.

What changes at this level

Level expectations

Each level is independent but clearly deeper than the last. This panel makes the jump explicit.

Assessment intent

Applied

Scenario based evaluation and pipeline decisions, including drift and governance basics.

Style

scenario

12 questions

Pass standard

Coming next

Not externally certified

▸Evidence you can save (CPD friendly)

An evaluation plan for one real decision: metrics, thresholds, and what errors cost you.
A small prompt or RAG workflow note: inputs, guardrails, tests, and a red-team example.
A monitoring checklist: drift signals, quality sampling, and a clear rollback or disable plan.

CPD timing

Applied time breakdown

Defensible timing based on page content: reading, labs, checkpoints, and reflection.

Reading

50m

7,422 words × 1.3

Practice

150m

10 × 15m

Checkpoints

25m

5 × 5m

Reflection

40m

5 × 8m

Estimated total

4h 25m

Based on page content

Claimed hours

Includes reattempts + capstone

CPD tracking

Fixed hours for this level are 4. Timed assessment time is included once on pass.

View in My CPD

Pricing and CPD Sign in to sync progress and unlock assessments

Progress minutes

0.0 hours

Learning objectives

What you will be able to do

1. Explain the training loop in plain language and diagnose common failure modes like overfitting and leakage.
2. Design a basic evaluation plan (metrics + thresholds) that matches a real-world decision and risk profile.
3. Describe how data quality, features, and representation affect model behaviour in production.
4. Select an appropriate deployment pattern (batch, real-time, retrieval-augmented) and explain the trade-offs.

What changes at this level

Level expectations

Each level is independent but clearly deeper than the last. This panel makes the jump explicit.

Assessment intent

Applied

Scenario based evaluation and pipeline decisions, including drift and governance basics.

Style

scenario

12 questions

Pass standard

Coming next

Not externally certified

▸Evidence you can save (CPD friendly)

An evaluation plan for one real decision: metrics, thresholds, and what errors cost you.
A small prompt or RAG workflow note: inputs, guardrails, tests, and a red-team example.
A monitoring checklist: drift signals, quality sampling, and a clear rollback or disable plan.

Learning contract

Applied outcomes

About 4 hours

Read the explanation first, then use the tools to test the idea. Skip any tool that is not useful for your goal.

Explain the training loop in plain language and diagnose common failure modes like overfitting and leakage.
Design a basic evaluation plan (metrics + thresholds) that matches a real-world decision and risk profile.
Describe how data quality, features, and representation affect model behaviour in production.
Select an appropriate deployment pattern (batch, real-time, retrieval-augmented) and explain the trade-offs.

Loading content...

Next step

Practise this level, then move on

I recommend you use the practice assessment for Applied to test your understanding and write a short reflection. Timed assessments are being prepared for this track.

Practice

Assessment

No timer

Pace

Reflection

Evidence

Practice assessment

Start the practice assessment for Applied

It is designed for confidence and evidence, and you can retry as often as you need.

Start practice assessment

The timed assessment for this level is being prepared. Use the practice assessment and labs until it is ready.

You can complete the course while signed out, and your progress saves in this browser. Sign in before assessments so your pass record is attached to your account.

Courses and assessments are free. There is no paywall for the learning path, practice questions, or formal assessments. Optional donations support hosting, maintenance, and ongoing updates.

During timed assessments, copy and the context menu are restricted to reduce casual cheating. Passed assessments are recorded in your account as evidence.

Course materials are protected by intellectual property rights.View terms

Loading lesson...

Core path

Your path through this level

Progress saves in this browser and syncs after you sign in

Completed

0 of 5

You will be able to

Explain the training loop in plain language and diagnose common failure modes like overfitting and leakage.
Design a basic evaluation plan (metrics + thresholds) that matches a real-world decision and risk profile.
Describe how data quality, features, and representation affect model behaviour in production.

Pick up where you left off

Training loop shape Features control behaviour Evaluation is evidence Deployment pattern Governance is enforcement

Optional

Full module map

Use this if you want the shape of the level before you start

Show

Module 2.1Training loop shape

Training is a loop: predict, measure error, update parameters, repeat with discipline.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain training loop shape in your own words and apply it to a realistic scenario.
Training is a loop: predict, measure error, update parameters, repeat with discipline.
Check the assumption "Loss matches the decision" and explain what changes if it is false.
Check the assumption "Validation is separate" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Overfitting. The model learns the training set, not the task. It looks strong and fails on new data.
Leakage. Information from the future sneaks into inputs. The score rises and trust collapses.

Module 2.2Features control behaviour

Representation choices decide what the model can notice and what it will ignore.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain features control behaviour in your own words and apply it to a realistic scenario.
Representation choices decide what the model can notice and what it will ignore.
Check the assumption "Meaning is preserved" and explain what changes if it is false.
Check the assumption "Bias is examined" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Proxy features. A feature can stand in for a sensitive attribute even if you never included it directly.
Schema drift. A feature changes meaning after a release. The model keeps working and starts guessing.

Module 2.3Evaluation is evidence

Evaluation is not one metric. It is evidence that the system is safe enough for the decision.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain evaluation is evidence in your own words and apply it to a realistic scenario.
Evaluation is not one metric. It is evidence that the system is safe enough for the decision.
Check the assumption "Metrics match harms" and explain what changes if it is false.
Check the assumption "You inspect failures" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Good average, bad edge cases. The average score can hide failures that matter most.
Threshold optimism. Thresholds set too aggressively create a system that looks good on paper and harms users.

Module 2.4Deployment pattern

Deployment is choosing a pattern that matches latency, cost, and safety.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain deployment pattern in your own words and apply it to a realistic scenario.
Deployment is choosing a pattern that matches latency, cost, and safety.
Check the assumption "Latency budget is known" and explain what changes if it is false.
Check the assumption "Costs are measured" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Hidden compute cost. Costs creep through retries, larger prompts, and wider retrieval. Measure and cap.
Missing fallbacks. When services fail, users still need a next step. Design for partial failure.

Module 2.5Governance is enforcement

Governance works when it is enforced by the system and measured by evidence.

Open

Prerequisites

Foundations-level vocabulary and concepts
Confidence with basic diagrams and section terminology

Outcomes

Explain governance is enforcement in your own words and apply it to a realistic scenario.
Governance works when it is enforced by the system and measured by evidence.
Check the assumption "Owners exist" and explain what changes if it is false.
Check the assumption "Evidence is reviewable" and explain what changes if it is false.

Practice

Work through one scenario and justify the decision with evidence
Compare two options and name the trade-off clearly

Artefact and failure modes

A one-page decision note with assumption, evidence, and chosen action
Policy only in text. A policy that is not enforced is a wish.
No change control. If changes are not tracked, you cannot explain failures or defend outcomes.

Optional

Planning and evidence

Objectives, timing, and CPD tracking

Show

Learning objectives

What you will be able to do

1. Explain the training loop in plain language and diagnose common failure modes like overfitting and leakage.
2. Design a basic evaluation plan (metrics + thresholds) that matches a real-world decision and risk profile.
3. Describe how data quality, features, and representation affect model behaviour in production.
4. Select an appropriate deployment pattern (batch, real-time, retrieval-augmented) and explain the trade-offs.

What changes at this level

Level expectations

Each level is independent but clearly deeper than the last. This panel makes the jump explicit.

Assessment intent

Applied

Scenario based evaluation and pipeline decisions, including drift and governance basics.

Style

scenario

12 questions

Pass standard

Coming next

Not externally certified

▸Evidence you can save (CPD friendly)

An evaluation plan for one real decision: metrics, thresholds, and what errors cost you.
A small prompt or RAG workflow note: inputs, guardrails, tests, and a red-team example.
A monitoring checklist: drift signals, quality sampling, and a clear rollback or disable plan.

CPD timing

Applied time breakdown

Defensible timing based on page content: reading, labs, checkpoints, and reflection.

Reading

50m

7,422 words × 1.3

Practice

150m

10 × 15m

Checkpoints

25m

5 × 5m

Reflection

40m

5 × 8m

Estimated total

4h 25m

Based on page content

Claimed hours

Includes reattempts + capstone

CPD tracking

Fixed hours for this level are 4. Timed assessment time is included once on pass.

View in My CPD

Pricing and CPD Sign in to sync progress and unlock assessments

Progress minutes

0.0 hours

Learning objectives

What you will be able to do

1. Explain the training loop in plain language and diagnose common failure modes like overfitting and leakage.
2. Design a basic evaluation plan (metrics + thresholds) that matches a real-world decision and risk profile.
3. Describe how data quality, features, and representation affect model behaviour in production.
4. Select an appropriate deployment pattern (batch, real-time, retrieval-augmented) and explain the trade-offs.

What changes at this level

Level expectations

Each level is independent but clearly deeper than the last. This panel makes the jump explicit.

Assessment intent

Applied

Scenario based evaluation and pipeline decisions, including drift and governance basics.

Style

scenario

12 questions

Pass standard

Coming next

Not externally certified

▸Evidence you can save (CPD friendly)

An evaluation plan for one real decision: metrics, thresholds, and what errors cost you.
A small prompt or RAG workflow note: inputs, guardrails, tests, and a red-team example.
A monitoring checklist: drift signals, quality sampling, and a clear rollback or disable plan.

Learning contract

Applied outcomes

About 4 hours

Read the explanation first, then use the tools to test the idea. Skip any tool that is not useful for your goal.

Explain the training loop in plain language and diagnose common failure modes like overfitting and leakage.
Design a basic evaluation plan (metrics + thresholds) that matches a real-world decision and risk profile.
Describe how data quality, features, and representation affect model behaviour in production.
Select an appropriate deployment pattern (batch, real-time, retrieval-augmented) and explain the trade-offs.

Loading content...

Next step

Practise this level, then move on

I recommend you use the practice assessment for Applied to test your understanding and write a short reflection. Timed assessments are being prepared for this track.

Practice

Assessment

No timer

Pace

Reflection

Evidence

Practice assessment

Start the practice assessment for Applied

It is designed for confidence and evidence, and you can retry as often as you need.

Start practice assessment

The timed assessment for this level is being prepared. Use the practice assessment and labs until it is ready.

You can complete the course while signed out, and your progress saves in this browser. Sign in before assessments so your pass record is attached to your account.

Courses and assessments are free. There is no paywall for the learning path, practice questions, or formal assessments. Optional donations support hosting, maintenance, and ongoing updates.

During timed assessments, copy and the context menu are restricted to reduce casual cheating. Passed assessments are recorded in your account as evidence.

Course materials are protected by intellectual property rights.View terms

Applied AI

What you will be able to do

Level expectations

Applied time breakdown

What you will be able to do

Level expectations

Applied outcomes

Practise this level, then move on

Start the practice assessment for Applied

Applied AI

What you will be able to do

Level expectations

Applied time breakdown

What you will be able to do

Level expectations

Applied outcomes

Practise this level, then move on

Start the practice assessment for Applied