AI-Assisted Full-Stack OA (Spring Boot + React)

A 90-minute take-home OA pairs one algorithm problem with two full-stack bug-fix tasks (frontend + backend) inside a packaged Spring Boot + React project, using an in-platform AI assistant of intentionally limited capability. Requirements Format: 7-day completion window; 90 minutes of active timer; proctored screen monitoring (no shared screen but extra-monitor detection); no sponsorship support is stated up front. Stack: candidate picks a backend (Spring Boot, Node, others) and a frontend (React, Angular). The OA ships a complete project skeleton in the chosen combination. Three tasks: Task 1 — Algorithm. A standard OA-grade coding problem (moderate difficulty, single function). Task 2 — Frontend bug fix. The bundled frontend has failing tests; candidate must locate the failing test, identify the relevant component / file, and repair the bug. An in-platform AI assistant is available. Task 3 — Backend bug fix. Same shape as Task 2 but in the backend service. The AI assistant is the same. Notes The dominant failure mode reported is wasted time discovering how the harness works. Test cases live in a separate tab and must be built and run before any error message appears. Spend the first five minutes mapping the project — pom.xml / package.json, test command, where test output surfaces — before opening any source file. The in-platform AI assistant is intentionally weaker than ChatGPT / Claude — candidates accustomed to strong AI tools report it being closer to GPT-3-class. Treat it as autocomplete + boilerplate generator, not as a reasoning partner. Read the failing test, form your own hypothesis, then ask the assistant to scaffold a fix; do not delegate the diagnosis. The bug-fix tasks are not algorithmic — they are debugging tasks in a real codebase. The skills graded are: reading an unfamiliar codebase fast, mapping a failing test to the file that produced it, and writing the smallest correct patch. The OA explicitly disallows sharing your own screen or using outside AI tools; proctoring detects extra monitors. Plan the workspace accordingly. Preparation Clone a small public Spring Boot + React starter and practice the loop: pick a failing test, find the file, fix the bug, re-run. Time yourself; aim for under ten minutes per cycle on familiar code. Rehearse Spring Boot fundamentals — bean lifecycle, @RestController routing, @Transactional semantics, profile-scoped config — since most planted bugs touch these surfaces. For React, the common planted bugs sit in state updates (useState stale closure), effect dependencies, and prop drilling. Refresh those patterns and the standard @testing-library/react failure messages. Treat the bundled AI assistant as a stand-in for autocomplete only. Practice the workflow of "diagnose first, prompt second" so a weak model does not become a critical-path dependency.