C++ Project Debugging with AI Tools

A larger C++ project is split into stages, each containing a logical bug, memory bug, or race condition. The candidate runs the program, observes output, debugs, and may use GPT / Gemini under screen share. Requirements The round gives a multi-file or multi-stage C++ project. Each stage has a different defect class: Incorrect output from flawed logic. Memory issue such as invalid lifetime, use-after-free, missing ownership rule, or bounds error. Race condition or thread-safety issue. Build / runtime failure that must be diagnosed from output. Workflow expected: Run the program and capture the failing output. Localize the failing stage. Explain the suspected defect before editing. Patch the bug. Re-run and verify. AI assistants may be allowed, but the screen is shared and the interviewer watches how prompts are written and how suggestions are validated. Notes Treat this as a debugging workflow interview, not a pure coding exercise. The signal is whether you can constrain the search space, build hypotheses, and verify a fix rather than pasting the whole codebase blindly. A strong AI-assisted workflow: Paste only the failing function, test output, and relevant data structures first. Ask for hypotheses, not final code. Inspect every suggested change before applying it. Add a targeted test or assertion after the fix. Explain why the bug was possible and how to prevent recurrence. For race conditions, be ready to discuss shared mutable state, lock granularity, atomics, condition variables, and deterministic repro strategies. Preparation Practice debugging small C++ programs with AddressSanitizer and ThreadSanitizer. Rehearse a concise screen-share narration: observation, hypothesis, experiment, result. If AI is allowed, keep prompts focused and show validation discipline.