Engineering manager technical fundamentals interview questions

As asked

Oracle's enterprise Java services often require tuning the JVM garbage collector to avoid long pauses. Explain the generational hypothesis, compare G1GC with ZGC, and describe how you would diagnose and reduce GC pause times in a service that must respond within 50 milliseconds.

Sample answer outline

A strong answer covers the generational hypothesis (most objects die young), how G1GC divides the heap into regions and predicts pause time, and how ZGC achieves sub-millisecond pauses by doing most work concurrently using load barriers. Diagnosis uses GC logs with -Xlog:gc*, heap dumps for retained objects, and JFR recordings. Tuning levers include heap sizing, region size in G1, and explicit tuning of young/old generation ratios.

Expect these follow-ups

• When would you choose GraalVM native image over a standard JVM deployment for an Oracle Cloud Function?
• How do you identify whether a pause is GC-related versus a safepoint stop caused by something else?

As asked

Flight computers on Falcon 9 and Starship run triple-redundant voting systems. Walk me through the software architecture patterns that make this work, and explain how you would test fault injection scenarios without access to the physical hardware.

Sample answer outline

A strong answer covers TMR (triple modular redundancy) at both the hardware and software level, the role of a voting arbiter that compares outputs across replicas and detects divergence, and how state synchronization between replicas is handled. For testing, the candidate should describe hardware-in-the-loop (HIL) simulators, software fault injection frameworks that kill or corrupt replicas programmatically, and formal verification for the most critical state machines. They should also mention deterministic execution requirements that make flight software difficult to test with ordinary unit tests.

Expect these follow-ups

• What happens in a 2-of-3 disagreement where one replica returns a different value? How does the arbiter decide?
• How do you manage software updates across three replicas without creating a window where they run different versions?

As asked

You are debugging a control system where a fault triggers once every few hundred runs with no obvious pattern. You cannot reproduce it in a lab environment consistently. Walk me through your approach.

Sample answer outline

A strong answer covers: adding deterministic logging at suspected race conditions without altering timing significantly (lock-free ring buffers for logging), using a reproducibility harness that records all inputs and replays them, using static and runtime race detectors (ThreadSanitizer), and hypothesis-driven reduction to isolate whether the bug is timing-related, data-related, or hardware-related. The candidate should mention that in aerospace you often cannot modify flight software to add debug instrumentation, so pre-planned observability and black-box recorders are essential.

Expect these follow-ups

• How do you know when you have fixed the bug versus when you just made it less frequent?
• What does your observability infrastructure look like before the bug is even reported?

As asked

OpenAI builds a significant amount of its own infrastructure, from training clusters to inference optimizations, but also uses third-party services. Walk me through how you decide whether to build something in-house versus buy or use an existing open-source solution for a given infrastructure problem.

Sample answer outline

A strong answer weighs differentiation (does this give us a competitive advantage?), cost at scale (vendor pricing vs engineering time), operational burden (who maintains it?), and strategic control (will the vendor's roadmap stay aligned with our needs?). They should give an example where they made this call and explain what data they used. For OpenAI specifically, custom inference kernels are a build; Stripe for payments is clearly a buy.

Expect these follow-ups

• When does it make sense to start with a vendor and then build in-house later?
• How do you account for the long-term cost of vendor lock-in in your analysis?