Anthropic interview questions

Showing 181–210 of 225 results

• Implement staircase printing and distributed mode/medianHard

  This set of problems evaluates implementation correctness and attention to formatting for a staircase-printing routine alongside distributed-algorithms reasoning for computing a global mode/median, testing algorithmic thinking, I/O handling, synchronization, and communication-cost trade-offs in the Coding & Algorithms domain.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Implement thread-safe blocking queueMedium

  This question evaluates competency in concurrent programming, synchronization primitives, and implementation of thread-safe bounded blocking queues within the Coding & Algorithms domain.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• In-Memory Key-Value Database with Nested TransactionsMedium

  Design an in-memory key-value store supporting nested transactions with begin, commit, rollback, get, set, and delete operations.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• LRU CacheMedium

  This question tests the ability to combine a hash map and a doubly linked list to implement an LRU cache with O(1) performance for both get and put operations.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Lead a Project Retrospective DiscussionHard

  Lead a project retrospective that explains the hardest challenge, its root cause, and the alternatives considered. Separate personal ownership from team effort, support outcomes with evidence, acknowledge limitations, and show how the lesson changed later decisions.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Make a Web Crawler Concurrent with Thread-Safe Visited StateMedium

  Convert a sequential crawler to bounded concurrency using atomic URL reservation, explicit retry state, and completion detection that accounts for active fetches.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Mode, Median and Sort Across 10 Nodes with Send/Receive While Minimizing TrafficMedium

  Integers are spread across 10 nodes that communicate only by sending and receiving messages. Write the per-node programs that compute the exact mode, the median and a globally sorted order while sending as little data as possible. It tests MapReduce-style aggregation, distributed selection, sample sort and message coordination.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Optimize MapReduce performanceMedium

  This question evaluates understanding of the MapReduce programming model and distributed batch processing at scale. It tests practical knowledge of parallel efficiency, network bottleneck analysis, and optimization techniques like combiners, partitioning, and speculative execution — core competencies assessed in machine learning and data engineering roles.

  AnthropicMachine Learning EngineerSystem DesignTechnical Screen
• Optimize a compute kernel with a simulatorHard

  This question evaluates performance engineering and low-level kernel optimization competencies, including profiling, data-layout and instruction-level techniques, correctness verification, and benchmark-driven measurement within the System Design domain.

  AnthropicSoftware EngineerSystem DesignOnsite
• Parallelize a Directory Walker With Thread-Safe Visited TrackingMedium

  Turn a sequential directory walker into a bounded-concurrency one, using an atomic reserve-before-expand step so two worker threads that reach the same directory at once never process it twice.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Path Resolution with Symbolic LinksMedium

  Implement a filesystem path resolution function that normalizes paths containing '.', '..', and symbolic links while detecting cycles.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Performance-Model a Sharded Matrix Multiply: FLOPs, Memory, Network and RooflineMedium

  Model the performance of a matrix multiplication by hand: count floating-point operations, memory bytes and network bytes, apply the roofline model to decide what bounds it, and compare sharding along each dimension across several accelerators. It tests arithmetic intensity, collective communication costs and quantitative reasoning.

  AnthropicSoftware EngineerML System DesignOnsite
• Prepare for a Recruiter Motivation ScreenMedium

  Tests your ability to articulate clear, credible motivations for joining an AI frontier lab, explain your current career inflection point, and justify your primary technical tooling choices during an initial recruiter screen.

  AnthropicSoftware EngineerBehavioral & LeadershipHR Screen
• Present a Technical Project to a Cross-Functional AudienceHard

  Tests your ability to present a complex engineering effort to a cross-functional technical panel, balancing architectural depth, consequential trade-offs, and rigorous post-launch reflection.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Present project and answer behavioralsMedium

  This question evaluates leadership and behavioral competencies alongside technical communication skills, including project ownership, system architecture articulation, trade-off reasoning, metrics-driven impact measurement, stakeholder management, and mentorship.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Reason About Delaying an AI Breakthrough Under Uncertain RiskHard

  Reason about an AI breakthrough under uncertain risk by separating research and deployment, weighing evidence and reversibility, and addressing competitive pressure consistently.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Repairing a Difficult Working RelationshipMedium

  Tests emotional intelligence, self-awareness, and constructive conflict resolution when collaborating with engineering peers or cross-functional partners under pressure.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Resolve a Package Dependency Graph with Concurrent FetchesMedium

  Discover a package's transitive dependency graph from a slow, flaky registry, first with a sequential BFS and then with a bounded pool of worker threads. Tests atomic visited-set reservation, explicit retry state, and correct termination detection with in-flight work.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Respond to a Hypothetical Loss of Employer Equity ValueMedium

  Tests how candidates navigate extreme organizational adversity, balance personal financial realities against mission alignment, and evaluate their long-term commitment and team responsibilities during a company crisis.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Resume Project Deep Dive and Prioritizing Too Many Tasks at OnceMedium

  A behavioral interview question in two parts: a deep dive into a project from your resume, where the interviewer probes design decisions, your own contribution and results, and a question about a time you had more tasks than you could finish. It tests technical depth, prioritization criteria and how you communicate trade-offs to stakeholders.

  AnthropicSoftware EngineerBehavioral & LeadershipOnsite
• Reuse Work Across Multiprocess Image PipelinesMedium

  Implement a thread-safe pipeline execution engine that caches deterministic intermediate prefix results, deduplicates concurrent in-flight computations, and manages a bounded LRU cache.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Review a Batched GPU Inference API Design Doc: Flaws, Throughput, Mixed-Model GPUsMedium

  Review a design document for a batched GPU inference API and find its flaws in load balancing, batching, caching, retries and admission control. Then compare theoretical and achievable throughput at 100 ms per batch and 10,000 QPS, and schedule a large and a small model on one 8-GPU pool.

  AnthropicSoftware EngineerSystem DesignOnsite
• Review an inference API design for scaleHard

  This question tests a candidate's ability to critically evaluate ML system designs for production-scale inference APIs, covering multi-tenancy, GPU resource constraints, and streaming token delivery. It assesses architectural reasoning across reliability, latency trade-offs, and capacity planning in the ML Systems domain.

  AnthropicSoftware EngineerML System DesignOnsite
• Review and Improve Four AI-Assisted TypeScript Pull RequestsHard

  Tests code review rigor, ability to audit AI-generated code for subtle regressions and security/type soundness, and hands-on refactoring with test-driven verification.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Review and Scale an Inference Service DesignMedium

  Tests the ability to architect a high-throughput ML inference service under strict latency SLAs, including dynamic batching, hardware capacity estimation, admission control, and multi-model routing.

  AnthropicSoftware EngineerML System DesignTechnical Screen
• Route and Batch Inference with Eight GPUsHard

  Design an inference routing, dynamic batching, and scheduling system across a fixed pool of eight 80GB GPUs serving both large and small language models under strict latency and memory constraints.

  AnthropicSoftware EngineerML System DesignTechnical Screen
• Same-Domain Web Crawl (BFS)Medium

  Implement a breadth-first search crawler that traverses pages starting from a seed URL, filtering for pages sharing the same domain and properly deduplicating visited URLs.

  AnthropicSoftware EngineerCoding & AlgorithmsOnsite
• Same-Host Web Crawler: Single-Threaded, Then Concurrent with a Thread-Safe Visited SetMedium

  Build a web crawler that starts from one URL, follows links only on the same hostname, strips URL fragments and fetches each page once, first single-threaded and then concurrently. It tests graph traversal, URL normalization, a thread-safe visited set and correct termination of concurrent work.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Scale a Thumbnail-and-Watermark Pipeline With Worker ProcessesMedium

  Design a batch pipeline that produces a resized thumbnail with a corner watermark for every uploaded photo, then extend it to run across a bounded pool of worker processes. Tests resource lifecycle, atomic output writes, per-item failure isolation, and what to pass between processes.

  AnthropicSoftware EngineerSoftware Engineering FundamentalsOnsite
• Scale crawler with thread poolHard

  This question evaluates system design and concurrent programming competencies, focusing on bounded thread-pool architecture, thread-safe frontier and deduplication structures, per-host politeness and in-flight limits, reliability mechanisms (timeouts, retries, backoff), and graceful shutdown and termination semantics.

  AnthropicSoftware EngineerSystem DesignTechnical Screen