Top 30 Interview Interview Questions and Answers

30 questions. Pick an answer, or open “Show the answer” to read it.

Answers are graded in your browser. Nothing is saved, and no XP or streak is earned here. The app keeps score.

1. Question 1 of 30

   In Kotlin, what does declaring a variable with val guarantee about the referenced object?

   • aAll properties of the object are automatically converted to read-only
   • bThe object becomes deeply immutable and thread-safe by default
   • cThe reference cannot be reassigned, though the object's contents may still mutate
   • dThe compiler prevents any modification to the object's internal state
   Show the answer

   Answer: c · The reference cannot be reassigned, though the object's contents may still mutate

   val only prevents reassigning the reference itself, so a val holding a MutableList can still be modified. The most tempting distractor confuses reference immutability with deep object immutability, which val does not enforce.

   Read the full bite: Explain val vs var in Kotlin and null safety risks

2. Question 2 of 30

   Which task best fits the SRE definition of toil rather than overhead or engineering project work?

   • aManually rerunning the same failed batch job every night by hand
   • bDesigning a new autoscaling system to handle future traffic
   • cAttending the weekly team planning meeting
   • dInterviewing candidates for an open SRE position
   Show the answer

   Answer: a · Manually rerunning the same failed batch job every night by hand

   Manually rerunning the same job nightly is manual, repetitive, automatable, and scales with the service, the hallmarks of toil. Designing a system is engineering; meetings and interviews are overhead.

   Read the full bite: What defines toil, with an example?

3. Question 3 of 30

   How should a healthy production ML lifecycle be structured from start to finish?

   • aAs a linear waterfall from algorithm selection to deployment with no iteration
   • bAs a process that starts with data ingestion and ends once the model is packaged for release
   • cAs an iterative loop starting with business problem framing and continuing through post-deployment monitoring
   • dAs a cycle focused primarily on training models and optimizing offline accuracy metrics
   Show the answer

   Answer: c · As an iterative loop starting with business problem framing and continuing through post-deployment monitoring

   The correct answer is B because the card describes the lifecycle as an end-to-end engineering process that begins with business goal definition and requires continuous monitoring and feedback loops after deployment. The most tempting distractor is D because treating packaging as the final step omits critical monitoring, retraining, and validation stages that keep a production model healthy.

   Read the full bite: Describe the key stages of a typical ML lifecycle

4. Question 4 of 30

   You write an async def FastAPI endpoint that calls requests.get. What is the main risk?

   • aFastAPI automatically moves the blocking call into a threadpool so it remains safe.
   • bIt runs correctly but is only slightly slower than using def for the same blocking call.
   • cThe event loop is blocked, freezing concurrent request handling until the call finishes.
   • dPython raises a SyntaxError because blocking libraries cannot be used inside async def.
   Show the answer

   Answer: c · The event loop is blocked, freezing concurrent request handling until the call finishes.

   The correct answer is C because calling a blocking library like requests inside async def stalls the event loop, stopping all other requests. The most tempting distractor is A because beginners often assume FastAPI magically threadpools any blocking code, but only def endpoints are run in a threadpool.

   Read the full bite: What is the difference between def and async def in Python and FastAPI?

5. Question 5 of 30

   If user is null, what happens when evaluating user?.name ?: "Guest"?

   • aIt throws a NullPointerException because safe calls crash on null receivers
   • bIt returns Guest without throwing an exception
   • cIt returns null because the safe call preserves the nullable type
   • dIt requires an explicit null check to compile safely
   Show the answer

   Answer: b · It returns Guest without throwing an exception

   The safe call operator ?. returns null when user is null, and the Elvis operator ?: then substitutes Guest as the fallback. Option C is tempting because ?. does preserve nullability, but it ignores that ?: immediately provides the default value.

   Read the full bite: What do the safe call and Elvis operators do in Kotlin?

6. Question 6 of 30

   Why should an internal SLO target be set stricter than the externally promised SLA?

   • aBecause customers cannot see SLOs, so they can be arbitrary
   • bBecause SLAs are measured monthly while SLOs are measured yearly
   • cBecause SLIs require the SLO and SLA to be numerically identical
   • dTo create a safety margin that triggers internal action before the contract is breached
   Show the answer

   Answer: d · To create a safety margin that triggers internal action before the contract is breached

   A stricter SLO gives early warning so the team reacts before violating the SLA and owing penalties. The other options misstate measurement windows, visibility, and the SLI relationship.

   Read the full bite: SLI vs SLO vs SLA: how do they relate?

7. Question 7 of 30

   Which statement correctly explains why an ID selector beats a class selector when they conflict?

   • aThe class rule wins because its CLASS column score of 1 is greater than the ID column score of 0
   • bThe class rule wins if it appears later, since source order overrides specificity when selectors differ
   • cThe ID rule wins because the specificity algorithm compares the ID column first, and 1-0-0 outranks 0-1-0
   • dThe ID rule wins because specificity is a single integer score and an ID is worth 100 points versus 10 for a class
   Show the answer

   Answer: c · The ID rule wins because the specificity algorithm compares the ID column first, and 1-0-0 outranks 0-1-0

   Specificity is a three-column value scored as ID-CLASS-TYPE, so an ID's 1-0-0 always beats a class's 0-1-0 before the CLASS column is ever evaluated. The most tempting distractor claims the later class wins, but source order only breaks ties when specificity is equal.

   Read the full bite: What color wins when an ID and class rule conflict?

8. Question 8 of 30

   Which statement best captures the relationship between SRE and DevOps?

   • aSRE is a prescriptive implementation of the broader DevOps philosophy
   • bSRE and DevOps are identical with no meaningful distinction
   • cThey are competing philosophies and an organization must choose one
   • dDevOps is simply the name for an SRE team's CI/CD pipeline
   Show the answer

   Answer: a · SRE is a prescriptive implementation of the broader DevOps philosophy

   SRE provides concrete practices, SLOs, error budgets, blameless postmortems, that implement the abstract DevOps principles. They are complementary, not competing, and they operate at different levels of abstraction.

   Read the full bite: How do SRE and DevOps relate?

9. Question 9 of 30

   When building an async execution-time decorator for FastAPI, why is omitting functools.wraps on the inner wrapper considered a red flag?

   • aIt causes the wrapper to run synchronously, blocking the event loop during high-concurrency requests.
   • bIt strips the original function's metadata, which breaks FastAPI's OpenAPI schema generation and dependency injection.
   • cIt makes FastAPI treat the endpoint as synchronous, offloading it to a thread pool and skewing the timer.
   • dIt prevents the decorator from awaiting the original coroutine, leading to unhandled runtime warnings.
   Show the answer

   Answer: b · It strips the original function's metadata, which breaks FastAPI's OpenAPI schema generation and dependency injection.

   functools.wraps preserves the original function's name, signature, and metadata, which FastAPI relies on to generate OpenAPI docs and resolve dependencies; omitting it exposes the wrapper's metadata instead. Distractor A is wrong because wraps has no effect on whether code runs synchronously or blocks the event loop.

   Read the full bite: Write an async decorator that logs execution time for FastAPI

10. Question 10 of 30

    When defining the first SLIs for a user-facing service, what should they primarily measure?

    • aWhatever metric is easiest to collect from the host
    • bAspects of the service as experienced by users, such as request success and latency
    • cThe number of deploys the team ships each week
    • dInternal server resource use like CPU and memory utilization
    Show the answer

    Answer: b · Aspects of the service as experienced by users, such as request success and latency

    Good SLIs reflect user-visible behavior, since the goal is to measure user happiness. CPU and deploy counts are internal signals that can look fine while users suffer.

    Read the full bite: How do you set SLOs for a service from scratch?

11. Question 11 of 30

    A class declares late final UserService repo and assigns it in initState. What happens if repo is read in build before that assignment?

    • aA LateInitializationError is thrown at the point of access
    • bThe compiler reports a definite-assignment error before the app runs
    • cThe app crashes with a NullPointerException because null safety is bypassed
    • dThe variable behaves as nullable and returns null until assigned
    Show the answer

    Answer: a · A LateInitializationError is thrown at the point of access

    late shifts definite-assignment checks from compile time to runtime, so accessing the field before assignment throws LateInitializationError. Distractor C is tempting because that is exactly the compile-time error you would receive if the field were not marked late.

    Read the full bite: What problem does late solve in Dart?

12. Question 12 of 30

    Which operation is valid on a variable of type unknown without first narrowing its type?

    • aAssigning it to a variable of type string
    • bAssigning a string value to the unknown variable
    • cPassing it as an argument to a function expecting a number
    • dCalling a method such as .toString() on it
    Show the answer

    Answer: b · Assigning a string value to the unknown variable

    Any value can be assigned to an unknown variable, but you cannot read properties, call methods, or assign it to a narrower type without narrowing first. Options A and D confuse unknown with any, which is assignable to all types, while B requires narrowing because the compiler cannot guarantee the method exists.

    Read the full bite: Explain the difference between any and unknown, and demonstrate type-safe narrowing

13. Question 13 of 30

    An error budget is fully spent early in the quarter. What is the most appropriate first response?

    • aIdentify which individuals caused incidents and hold them accountable
    • bPermanently ban all releases for the rest of the year
    • cIgnore it, since revenue features should always take priority
    • dInvoke the pre-agreed error budget policy and analyze what consumed the budget
    Show the answer

    Answer: d · Invoke the pre-agreed error budget policy and analyze what consumed the budget

    The budget is a pre-agreed signal that triggers a policy and a data-driven analysis of the burn. Blaming individuals, a permanent ban, or ignoring it all defeat the budget's purpose as an objective tradeoff tool.

    Read the full bite: Error budget exhausted early: what now?

14. Question 14 of 30

    What is the best way to prioritize which toil to automate first?

    • aAutomate whichever task the on-call engineer finds most irritating
    • bAutomate the task that is technically easiest regardless of its impact
    • cRank by return on investment, weighing frequency and time saved against automation effort and risk
    • dCount how many scripts the team can write in a sprint
    Show the answer

    Answer: c · Rank by return on investment, weighing frequency and time saved against automation effort and risk

    Prioritizing by ROI, frequency times time saved versus effort and risk, maximizes recovered engineering time. Irritation, ease alone, or script count are poor proxies for actual impact.

    Read the full bite: How do you find and eliminate toil systematically?

15. Question 15 of 30

    Which statement accurately describes how Dart closures handle captured variables from an enclosing function?

    • aThey copy primitive variables by value and objects by reference at creation time
    • bThey keep the enclosing function's stack frame alive until the closure is destroyed
    • cThey can only be created by anonymous functions, not by named nested functions
    • dThey capture the variable binding by reference in a heap-allocated environment
    Show the answer

    Answer: d · They capture the variable binding by reference in a heap-allocated environment

    Dart closures capture the variable binding by reference in the heap, so the inner function retains access even after the outer function returns and its stack frame is popped. This makes B incorrect because the enclosing function's stack frame does not remain on the call stack after it returns.

    Read the full bite: What is a Dart closure? Write a function that returns a function.

16. Question 16 of 30

    Why is demosaicing necessary after a Bayer sensor captures an image?

    • aBecause each photosite records only one color channel, leaving missing values to estimate.
    • bBecause the raw data requires white balance and gamma correction before display.
    • cBecause the sensor merges multiple short exposures that must be aligned.
    • dBecause the sensor outputs full RGB per pixel in a compressed mosaic layout.
    Show the answer

    Answer: a · Because each photosite records only one color channel, leaving missing values to estimate.

    Demosaicing is required because every photosite measures only a single color channel, so the missing two channels must be interpolated from neighbors. Option D represents the common misconception that Bayer pixels already contain complete RGB data.

    Read the full bite: How does a Bayer filter capture color and what is demosaicing?

17. Question 17 of 30

    Which getProperty signature both rejects invalid keys at compile time and returns the exact type of the requested property?

    • aA single generic T with the key typed as string and the return type set to any
    • bTwo generics T and K where K is unconstrained and the return type is T[K]
    • cTwo generics T and K where K extends keyof T and the return type is T[K]
    • dA single generic T with the key typed as keyof T and returning T[keyof T]
    Show the answer

    Answer: c · Two generics T and K where K extends keyof T and the return type is T[K]

    Option C is correct because constraining K with extends keyof T limits keys to valid properties on T, while T[K] preserves the exact type of the accessed property. Option D is tempting because keyof T does restrict keys, but T[keyof T] produces a union of all property types rather than the specific one for the key passed.

    Read the full bite: Create a generic getProperty using generics and keyof

18. Question 18 of 30

    Why is blamelessness essential to a postmortem's effectiveness at improving reliability?

    • aIt lets the team avoid documenting any action items
    • bIt speeds up writing the report by skipping the root-cause section
    • cIt creates psychological safety so engineers disclose full, honest details needed to fix systemic causes
    • dIt guarantees the same incident can never technically recur
    Show the answer

    Answer: c · It creates psychological safety so engineers disclose full, honest details needed to fix systemic causes

    Blamelessness removes fear of punishment so people share complete information, and you can only fix what you fully understand. It does not skip root cause or action items, nor magically prevent recurrence by itself.

    Read the full bite: What makes a blameless postmortem effective?

19. Question 19 of 30

    In a specificity conflict between `#header` and a selector with eleven classes, which statement is true?

    • aThe `#header` selector wins because the ID column outweighs any number of classes
    • bThe eleven-class selector wins because eleven simple selectors outweigh one
    • cThe `#header` selector wins because inline styles are included in the selector score
    • dThe eleven-class selector wins because its specificity sums to a larger integer than the ID
    Show the answer

    Answer: a · The `#header` selector wins because the ID column outweighs any number of classes

    The card explains that specificity is a tuple where the leftmost nonzero column wins, so one ID always beats any number of classes. Option D is tempting because it reflects the common error of collapsing the tuple into a single integer.

    Read the full bite: How is CSS specificity calculated for a complex selector?

20. Question 20 of 30

    Why does self-attention use three separate learned projections of the same input rather than the raw embeddings directly?

    • aIt allows the model to learn which token features to use for matching versus which to pass forward as content
    • bBecause the weight matrices are predefined linguistic features that do not require training
    • cBecause the dot product of raw embeddings would be too large and cause softmax saturation
    • dSo that the query can come from the decoder while key and value come from the encoder
    Show the answer

    Answer: a · It allows the model to learn which token features to use for matching versus which to pass forward as content

    The correct answer reflects that learned projections decouple the matching process from content retrieval, letting the model decide which aspects of a token to use for scoring and which to propagate forward. The distractor describing a decoder query with encoder key and value defines cross-attention, not self-attention, where all three matrices are derived from the same input sequence.

    Read the full bite: Explain Q, K, and V matrices in self-attention

21. Question 21 of 30

    In Swift, what is a key advantage of providing default behavior through a protocol extension rather than an abstract base class?

    • aIt allows conforming types to dynamically override defaults at runtime through subclassing.
    • bProtocol extensions can add stored properties to all conforming types.
    • cIt requires conforming classes to use the objc attribute to access the default implementation.
    • dIt enables structs and enums to reuse behavior without being forced into an inheritance hierarchy.
    Show the answer

    Answer: d · It enables structs and enums to reuse behavior without being forced into an inheritance hierarchy.

    Protocol extensions let structs and enums gain shared behavior without being forced into an inheritance hierarchy, unlike abstract base classes. They cannot add stored properties, so any answer suggesting they extend state is incorrect.

    Read the full bite: Explain protocols and how extensions provide default implementations

22. Question 22 of 30

    Which approach best lets a risky feature launch despite a nearly exhausted error budget while upholding reliability?

    • aRefuse the launch outright until the next quarter's budget resets
    • bDisable monitoring during launch so the feature's errors do not count against the budget
    • cShip to all users immediately since the revenue upside is large
    • dRoll out behind a flag to a small canary, gate progression on live burn, and get explicit risk sign-off
    Show the answer

    Answer: d · Roll out behind a flag to a small canary, gate progression on live burn, and get explicit risk sign-off

    Canary plus flag plus burn-gated rollout and documented risk acceptance contains blast radius while enabling the business. A flat refusal, a full rollout, or hiding errors all abandon reliability discipline.

    Read the full bite: Risky launch with a near-empty error budget?

23. Question 23 of 30

    A conversion metric drops suddenly with no recent deployments. Segmenting reveals an identical percentage loss across all devices, channels, and geographies. What does this pattern most strongly indicate?

    • aA genuine business regression such as a paused campaign or pricing change
    • bA temporary vendor processing delay that will self-correct within days
    • cA localized technical issue like a regional checkout bug
    • dA global instrumentation problem such as a broken tag or attribution change
    Show the answer

    Answer: d · A global instrumentation problem such as a broken tag or attribution change

    A uniform percentage loss across every dimension is the hallmark of a global instrumentation break such as a missing tag or attribution change. The most tempting distractor is a business regression, but jumping to product or marketing causes without first reconciling frontend events against backend transactions treats the metric as ground truth before validating data integrity.

    Read the full bite: Conversion metric dropped suddenly with no recent deployments; debug instrumentation causes

24. Question 24 of 30

    What fundamentally distinguishes the SRE response to a recurring high-volume alert from a traditional ops response?

    • aSRE treats it as a defect to automate or eliminate so effort scales sublinearly with load
    • bSRE responds to each alert faster but still handles it manually every time
    • cSRE buys more dashboards while ops uses fewer
    • dSRE always adds more on-call engineers to cover the volume
    Show the answer

    Answer: a · SRE treats it as a defect to automate or eliminate so effort scales sublinearly with load

    SRE applies software engineering to remove the recurring work entirely, breaking the link between load and headcount. Faster manual response, more dashboards, or more engineers are the linear-scaling ops pattern SRE avoids.

    Read the full bite: SRE vs traditional ops on a recurring alert?

25. Question 25 of 30

    When writing a swap function in Swift, what advantage does using a generic placeholder T provide over accepting parameters of type Any?

    • aGenerics use runtime boxing to share a single machine code implementation across all types
    • bGenerics automatically convert value types to reference types during the swap
    • cGenerics allow swapping two values of unrelated types in a single function call
    • dGenerics enforce that both parameters are the same concrete type and avoid runtime casting
    Show the answer

    Answer: d · Generics enforce that both parameters are the same concrete type and avoid runtime casting

    Generics preserve compile-time type information, ensuring both arguments share the same type and eliminating unsafe downcasting. Distractor A is tempting because Any permits heterogeneous values, but a generic swap<T> explicitly prevents mixing types, which is exactly why it is type-safe.

    Read the full bite: What are Swift generics, why useful, and write a swap function?

26. Question 26 of 30

    An upstream service breaches its SLO solely because a downstream dependency had an outage. How should a well-designed error budget policy handle the burn?

    • aCharge the burn to the upstream team since the breach showed up in their SLO
    • bIgnore the burn because cross-service failures are unavoidable
    • cAttribute the burn to the downstream service that caused the failure
    • dSplit the burn evenly across every team in the dependency graph
    Show the answer

    Answer: c · Attribute the burn to the downstream service that caused the failure

    Correct attribution charges the responsible downstream team, creating proper incentives and shielding the upstream victim. Charging the upstream team, splitting blindly, or ignoring it all distort accountability.

    Read the full bite: Error budget policy across dependent microservices?

27. Question 27 of 30

    In a distributed web service, which telemetry type is best for pinpointing which downstream service is adding latency to a single slow request?

    • aLogs alone, because they record every event with full context
    • bAggregated metrics, because they show overall latency percentiles
    • cNone; you must guess from the dashboard trend
    • dDistributed traces, because they follow one request's spans across services
    Show the answer

    Answer: d · Distributed traces, because they follow one request's spans across services

    Traces follow a single request across services and attribute time to each span, localizing the slow hop. Metrics show aggregate trends and logs give per-event detail, but neither maps one request's cross-service path like a trace.

    Read the full bite: Explain the three pillars of observability

28. Question 28 of 30

    When designing a pipeline to discover unknown pain-point categories from thousands of unstructured reviews, which sequence best ensures valid grouping and reliable severity ranking?

    • aFeed raw reviews directly into an LLM for summarization, then rank the most frequent terms as the top pain points.
    • bTokenize and remove stopwords, then run sentiment analysis on the full corpus before grouping reviews into predefined product areas.
    • cDeduplicate and normalize the corpus, cluster to discover themes, then apply sentiment analysis within each cluster to rank by severity.
    • dApply supervised classification with a predefined taxonomy, then count negative keywords per class to rank issues.
    Show the answer

    Answer: c · Deduplicate and normalize the corpus, cluster to discover themes, then apply sentiment analysis within each cluster to rank by severity.

    The correct sequence matches the card's recommended lifecycle: preprocess to remove noise and duplicates, use unsupervised clustering to discover emergent themes since categories are unknown, and score sentiment within each cluster to rank pain points by frequency and severity. Option A is a tempting distractor because LLMs are popular, but the card flags jumping straight to summarization without cleaning as a red flag that yields unreliable, unvalidated output.

    Read the full bite: Outline an NLP pipeline to categorize reviews and identify pain points

29. Question 29 of 30

    Which set correctly lists the Four Golden Signals for monitoring a user-facing system?

    • aCPU, memory, disk, and network utilization
    • bLatency, traffic, errors, and saturation
    • cAvailability, durability, throughput, and cost
    • dLogs, metrics, traces, and dashboards
    Show the answer

    Answer: b · Latency, traffic, errors, and saturation

    The Four Golden Signals are latency, traffic, errors, and saturation, focused on user experience and capacity. CPU and memory are resource metrics; logs and traces are observability pillars, not the golden signals.

    Read the full bite: What are the Four Golden Signals?

30. Question 30 of 30

    When initializing UI handlers on a page with external images, which distinction between DOMContentLoaded and window.load matters most?

    • aDOMContentLoaded always blocks until all stylesheets finish downloading, so load is safer for initialization
    • bwindow.load fires once the HTML parser finishes, while DOMContentLoaded waits for every stylesheet and image
    • cThe two events are interchangeable today, though load provides better backward compatibility
    • dDOMContentLoaded fires after deferred scripts run but does not wait for images, letting handlers bind earlier
    Show the answer

    Answer: d · DOMContentLoaded fires after deferred scripts run but does not wait for images, letting handlers bind earlier

    DOMContentLoaded fires after HTML parsing and deferred scripts complete, so you can bind UI handlers before heavy assets like images finish, whereas window.load waits for all subresources. The most tempting distractor reverses the two events: window.load is the one that waits for every asset, not DOMContentLoaded.

    Read the full bite: Describe the difference between DOMContentLoaded and window.load