roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Gets you ready for QA and test-automation engineer interviews, from testing theory to tooling rounds. Covers quality fundamentals, UI and API automation, performance testing, databases, and CI/CD.
6463 of 6463 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 6463 new. 0 cut, outside your plan.
A QA loop usually opens with a screen and then a conversation with no code in it: a login form or a search box goes on the table, and the interviewer listens to how you take it apart. A weak answer names inputs to try; a strong one splits them into classes, tests the edges of each (empty, the maximum, one past it), asks what the requirement leaves unsaid, and says which cases deserve automation and which are cheaper to explore by hand. A hands-on round follows with UI or API tests to write, plus a short coding problem when the posting leans toward engineering in test; a behavioral conversation, often about a bug that shipped anyway, closes most loops. The automation questions are about why tests lie. What does Playwright wait for on its own before a click, and what is still yours to wait for? Why can a Cypress command's result not be stored in a variable like a return value? What does a test stop proving when it logs in through the API and reuses the session cookie? Why does a load test report flattering latency when each virtual user waits for its previous response? Seniority shows in what you are asked to own: a feature and its cases early on, a whole suite from middle level up (which layer a check belongs in, what to do with a test that fails with no code change), and at senior level the framework and the process around it, including what you would leave unautomated.
[Testing theory](/topics/found-testing-quality) opens the plan, since a tool can only automate a test someone has already designed, and every later module is measured against it. Accessibility sits beside it, and returns as assertions inside component tests. Delivery process follows, because a tester's say over a release is exercised in refinement and in the [definition of done](/topics/found-scrum-dod), not in a test plan written afterwards. The protocol then comes before any tool that speaks it. [HTTP](/topics/proto-http), REST and OpenAPI are where an argument about a bug gets settled: whether a `200` carrying an error message is a defect, whether a retried `PUT` may change anything twice, whether a response honors its spec. UI work starts with design rather than drivers: [end-to-end patterns](/topics/fe-testing-e2e-patterns) such as selector strategy and waiting are framework-neutral, so what sets [Selenium](/topics/dev-selenium), [Playwright](/topics/dev-playwright) and [Cypress](/topics/dev-cypress) apart afterwards is where each one sits: a driver behind the WebDriver protocol, a direct connection to the browser, code running inside the page. Device clouds run those suites across real browsers, Cucumber puts readable scenarios over them, and [Appium](/topics/mob-appium) answers the same WebDriver commands against a native app on a phone, where locating an element and timing an interaction both get harder. On the API side the unit of a test grows: one request checked in Postman and run headless by its runner [Newman](/topics/dev-newman), then suites written in REST Assured or Karate, then the seam between services, where a stub server stands in for a dependency and [contract testing](/topics/dev-contract-testing) asks whether two services still agree. Performance testing puts the same endpoints under load with generators that differ mostly in how a test is written, from JMeter's GUI-built plans to [k6](/topics/dev-k6) scripts in JavaScript, and ends on [capacity planning](/topics/cloud-sre-practices-capacity), saturation seen from the operator's side. Test management closes the testing half by turning results into a release decision. Then the plan turns to the systems tests run in. [SQL](/topics/lang-sql) arrives with keys, constraints and schema design, because an API call can report success and still leave an orphaned row or a duplicate the schema should have refused. Source control and Linux are daily ground; [CI/CD](/topics/cloud-cicd-concepts) and Docker decide when a suite runs and against what, and metrics, logs and Grafana dashboards show whether a red run was the product or the environment. Security testing pairs secure-coding principles with [ZAP](/topics/sec-owasp-zap), a scanner a test team can run in its own pipeline. The close serves the remaining rounds: AI-assisted coding, the [algorithms](/topics/found-dsa) behind the coding problem, and the [behavioral round](/topics/found-behavioral).
Testers coming from manual work know the opening modules well and are tempted to go straight to a tool. Hold off: tool questions come easier once HTTP and the end-to-end patterns are solid, and choosing by topics for study order keeps them in that sequence. Developers moving the other way can skim source control and pipelines; what interviewers push back on in their answers is a suite of happy paths with a high coverage figure, so the testing theory deserves their hours. Short on time, go deep on the UI tool the posting names and read only the comparison sections of the other two, since why one tool over another gets asked anyway. Level and importance then cut what remains: a level range no wider than the posting's band, and only the must know end of importance. In the last days, a pass with study order on random rehearses a QA screen, which jumps from a status code to a sprint ceremony to a locator. The plan teaches no general-purpose programming language; the automation modules assume you already write the one your framework uses. If that language is Java and the job leans toward engineering, the [Java SDET roadmap](/roadmaps/java-sdet) puts the language first and builds most of these testing modules on top of it.
A QA loop usually opens with a screen and then a conversation with no code in it: a login form or a search box goes on the table, and the interviewer listens to how you take it apart. A weak answer names inputs to try; a strong one splits them into classes, tests the edges of each (empty, the maximum, one past it), asks what the requirement leaves unsaid, and says which cases deserve automation and which are cheaper to explore by hand. A hands-on round follows with UI or API tests to write, plus a short coding problem when the posting leans toward engineering in test; a behavioral conversation, often about a bug that shipped anyway, closes most loops.
The automation questions are about why tests lie. What does Playwright wait for on its own before a click, and what is still yours to wait for? Why can a Cypress command's result not be stored in a variable like a return value? What does a test stop proving when it logs in through the API and reuses the session cookie? Why does a load test report flattering latency when each virtual user waits for its previous response?
Seniority shows in what you are asked to own: a feature and its cases early on, a whole suite from middle level up (which layer a check belongs in, what to do with a test that fails with no code change), and at senior level the framework and the process around it, including what you would leave unautomated.
Testing theory opens the plan, since a tool can only automate a test someone has already designed, and every later module is measured against it. Accessibility sits beside it, and returns as assertions inside component tests. Delivery process follows, because a tester's say over a release is exercised in refinement and in the definition of done, not in a test plan written afterwards.
The protocol then comes before any tool that speaks it. HTTP, REST and OpenAPI are where an argument about a bug gets settled: whether a 200 carrying an error message is a defect, whether a retried PUT may change anything twice, whether a response honors its spec. UI work starts with design rather than drivers: end-to-end patterns such as selector strategy and waiting are framework-neutral, so what sets Selenium, Playwright and Cypress apart afterwards is where each one sits: a driver behind the WebDriver protocol, a direct connection to the browser, code running inside the page. Device clouds run those suites across real browsers, Cucumber puts readable scenarios over them, and Appium answers the same WebDriver commands against a native app on a phone, where locating an element and timing an interaction both get harder.
On the API side the unit of a test grows: one request checked in Postman and run headless by its runner Newman, then suites written in REST Assured or Karate, then the seam between services, where a stub server stands in for a dependency and contract testing asks whether two services still agree. Performance testing puts the same endpoints under load with generators that differ mostly in how a test is written, from JMeter's GUI-built plans to k6 scripts in JavaScript, and ends on capacity planning, saturation seen from the operator's side. Test management closes the testing half by turning results into a release decision.
Then the plan turns to the systems tests run in. SQL arrives with keys, constraints and schema design, because an API call can report success and still leave an orphaned row or a duplicate the schema should have refused. Source control and Linux are daily ground; CI/CD and Docker decide when a suite runs and against what, and metrics, logs and Grafana dashboards show whether a red run was the product or the environment. Security testing pairs secure-coding principles with ZAP, a scanner a test team can run in its own pipeline. The close serves the remaining rounds: AI-assisted coding, the algorithms behind the coding problem, and the behavioral round.
Testers coming from manual work know the opening modules well and are tempted to go straight to a tool. Hold off: tool questions come easier once HTTP and the end-to-end patterns are solid, and choosing by topics for study order keeps them in that sequence. Developers moving the other way can skim source control and pipelines; what interviewers push back on in their answers is a suite of happy paths with a high coverage figure, so the testing theory deserves their hours.
Short on time, go deep on the UI tool the posting names and read only the comparison sections of the other two, since why one tool over another gets asked anyway. Level and importance then cut what remains: a level range no wider than the posting's band, and only the must know end of importance. In the last days, a pass with study order on random rehearses a QA screen, which jumps from a status code to a sprint ceremony to a locator.
The plan teaches no general-purpose programming language; the automation modules assume you already write the one your framework uses. If that language is Java and the job leans toward engineering, the Java SDET roadmap puts the language first and builds most of these testing modules on top of it.