1 · can start
Same Prompt, Different Test Suite
~0.8 h
Generative AI for Software Testing: A Practitioner's Course
A hands-on course on using large language models across the test process: prompting for test analysis and design, evaluating what the model gives back, containing its risks, and rolling it out in an organisation that handles client data. Built around a professional-services engagement, where confidentiality and documented review are not optional. Assumes working knowledge of test design, test levels and defect management.
Material and exam: English · your assistant leads the lesson
How it works
Your assistant takes the course material from our server and leads the lesson in your language.
The exam is taken on the site, without an assistant. Exam questions are not translated: this is decided for every course.
Path through the course
· 21 unitssolid line — the way is open · dashed — closed for nowView: map · steps
2 · closed
The Model You Picked Costs You Twice
~0.7 h
opens after “Same Prompt, Different Test Suite”
3 · closed
The Screenshot Your Model Cannot Read
~0.7 h
opens after “The Model You Picked Costs You Twice”
4 · closed
"Make Me Some Test Cases"
~1.1 h
opens after “The Screenshot Your Model Cannot Read”
5 · closed
Three Techniques, and Where Each Breaks
~1.1 h
opens after “"Make Me Some Test Cases"”
6 · closed
Finding the Holes in the Requirement
~1 h
opens after “Three Techniques, and Where Each Breaks”
7 · closed
From Acceptance Criteria to Test Cases
~1 h
opens after “Finding the Holes in the Requirement”
8 · closed
Gherkin at Scale
~0.9 h
opens after “From Acceptance Criteria to Test Cases”
9 · closed
The Regression Suite That Heals Itself
~0.9 h
opens after “Gherkin at Scale”
10 · closed
Reading a Failed Nightly Run
~0.8 h
opens after “The Regression Suite That Heals Itself”
11 · closed
Numbers the Partner Will Read
~0.7 h
opens after “Reading a Failed Nightly Run”
12 · closed
Choose the Technique Before You Write
~0.8 h
opens after “Numbers the Partner Will Read”
13 · closed
Proving the Output Is Good Enough
~0.9 h
opens after “Choose the Technique Before You Write”
14 · closed
It Sounded Right and It Was Wrong
~0.9 h
opens after “Proving the Output Is Good Enough”
15 · closed
Turning the Temperature Down
~0.7 h
opens after “It Sounded Right and It Was Wrong”
16 · closed
Client Data Does Not Leave the Firm
~1 h
opens after “Turning the Temperature Down”
17 · closed
Who Signs Off That This Was Legal
~0.9 h
opens after “Client Data Does Not Leave the Firm”
18 · closed
The Model Has Never Seen Your Test Cases
~1.1 h
opens after “Who Signs Off That This Was Legal”
19 · closed
The Agent That Filed Forty Bad Defects
~0.8 h
opens after “The Model Has Never Seen Your Test Cases”
20 · closed
When Prompting Is Not Enough
~0.9 h
opens after “The Agent That Filed Forty Bad Defects”
21 · closed
Shadow AI in an Audit Firm
~0.9 h
opens after “When Prompting Is Not Enough”
final · in the browser, without the assistant
Exam
closed · enrol on the course
The path in words
- Stage 1 · Same Prompt, Different Test Suite — can start; ~0.8 h
- Stage 2 · The Model You Picked Costs You Twice — closed; opens after “Same Prompt, Different Test Suite”; ~0.7 h
- Stage 3 · The Screenshot Your Model Cannot Read — closed; opens after “The Model You Picked Costs You Twice”; ~0.7 h
- Stage 4 · "Make Me Some Test Cases" — closed; opens after “The Screenshot Your Model Cannot Read”; ~1.1 h
- Stage 5 · Three Techniques, and Where Each Breaks — closed; opens after “"Make Me Some Test Cases"”; ~1.1 h
- Stage 6 · Finding the Holes in the Requirement — closed; opens after “Three Techniques, and Where Each Breaks”; ~1 h
- Stage 7 · From Acceptance Criteria to Test Cases — closed; opens after “Finding the Holes in the Requirement”; ~1 h
- Stage 8 · Gherkin at Scale — closed; opens after “From Acceptance Criteria to Test Cases”; ~0.9 h
- Stage 9 · The Regression Suite That Heals Itself — closed; opens after “Gherkin at Scale”; ~0.9 h
- Stage 10 · Reading a Failed Nightly Run — closed; opens after “The Regression Suite That Heals Itself”; ~0.8 h
- Stage 11 · Numbers the Partner Will Read — closed; opens after “Reading a Failed Nightly Run”; ~0.7 h
- Stage 12 · Choose the Technique Before You Write — closed; opens after “Numbers the Partner Will Read”; ~0.8 h
- Stage 13 · Proving the Output Is Good Enough — closed; opens after “Choose the Technique Before You Write”; ~0.9 h
- Stage 14 · It Sounded Right and It Was Wrong — closed; opens after “Proving the Output Is Good Enough”; ~0.9 h
- Stage 15 · Turning the Temperature Down — closed; opens after “It Sounded Right and It Was Wrong”; ~0.7 h
- Stage 16 · Client Data Does Not Leave the Firm — closed; opens after “Turning the Temperature Down”; ~1 h
- Stage 17 · Who Signs Off That This Was Legal — closed; opens after “Client Data Does Not Leave the Firm”; ~0.9 h
- Stage 18 · The Model Has Never Seen Your Test Cases — closed; opens after “Who Signs Off That This Was Legal”; ~1.1 h
- Stage 19 · The Agent That Filed Forty Bad Defects — closed; opens after “The Model Has Never Seen Your Test Cases”; ~0.8 h
- Stage 20 · When Prompting Is Not Enough — closed; opens after “The Agent That Filed Forty Bad Defects”; ~0.9 h
- Stage 21 · Shadow AI in an Audit Firm — closed; opens after “When Prompting Is Not Enough”; ~0.9 h
Programme · 21 units
| 1 | Same Prompt, Different Test Suite Explain why an LLM returns different output for identical input | ~0.8 h | can start |
| 2 | The Model You Picked Costs You Twice Distinguish the foundation, instruction-tuned and reasoning stages of LLM training | ~0.7 h | closed |
| 3 | The Screenshot Your Model Cannot Read Write and execute a prompt combining a wireframe and a user story for a test task | ~0.7 h | closed |
| 4 | "Make Me Some Test Cases" Give examples of the six components of a structured prompt for a test task | ~1.1 h | closed |
| 5 | Three Techniques, and Where Each Breaks Differentiate prompt chaining, few-shot prompting and meta prompting | ~1.1 h | closed |
| 6 | Finding the Holes in the Requirement Apply generative AI to identify defects in the test basis | ~1 h | closed |
| 7 | From Acceptance Criteria to Test Cases Apply generative AI to generate test cases from acceptance criteria | ~1 h | closed |
| 8 | Gherkin at Scale Use few-shot prompting to produce output in a fixed house format | ~0.9 h | closed |
| 9 | The Regression Suite That Heals Itself Apply few-shot prompting to generate keyword-driven test scripts | ~0.9 h | closed |
| 10 | Reading a Failed Nightly Run Apply structured prompts to analyse a regression test report | ~0.8 h | closed |
| 11 | Numbers the Partner Will Read Apply generative AI to test monitoring and test control tasks | ~0.7 h | closed |
| 12 | Choose the Technique Before You Write Select an appropriate prompting technique for a given context and test task | ~0.8 h | closed |
| 13 | Proving the Output Is Good Enough Understand the metrics for evaluating generative AI results on test tasks | ~0.9 h | closed |
| 14 | It Sounded Right and It Was Wrong Recall the definitions of hallucination, reasoning error and bias | ~0.9 h | closed |
| 15 | Turning the Temperature Down Recall mitigation techniques for non-deterministic LLM behaviour | ~0.7 h | closed |
| 16 | Client Data Does Not Leave the Firm Explain the key data privacy and security risks of using generative AI in testing | ~1 h | closed |
| 17 | Who Signs Off That This Was Legal Recall the AI regulations, standards and best-practice frameworks relevant to testing | ~0.9 h | closed |
| 18 | The Model Has Never Seen Your Test Cases Explain the architectural components of LLM-powered test infrastructure | ~1.1 h | closed |
| 19 | The Agent That Filed Forty Bad Defects Explain the role of LLM-powered agents in automating test process tasks | ~0.8 h | closed |
| 20 | When Prompting Is Not Enough Explain fine-tuning of language models for test tasks and when it is justified | ~0.9 h | closed |
| 21 | Shadow AI in an Audit Firm Recall the risks of shadow AI in an organisation handling client data | ~0.9 h | closed |