Anthropic

A recruiter call, a leveled CodeSignal assessment and a live practical screen, then about five onsite rounds over two days: coding and system design usually come first, and a weak first day can cancel the second (hiring manager, project deep dive, culture). Two coworker references are checked before the offer.

• 

  Image transform pipeline with a wall-clock budgetCodingApply a fixed sequence of image transforms (grayscale, flip, scale, blur, rotate) to every image in a directory, then speed the batch up with threads or processes to hit a…

  Last asked Oct 2026Were you asked this too?
• 

  Concurrent same-host web crawlerCodingImplement a web crawler that finds every page reachable from a start URL on the same host, and fetches pages concurrently. You are given a blocking helper: Write: python def…

  Last asked Oct 2026Were you asked this too?
• 

  Design a distributed AI model downloaderAI systemGet a new multi-hundred-gigabyte model checkpoint onto thousands of GPU hosts as fast as possible without overloading the storage it comes from.

  Last asked Oct 2026Were you asked this too?
• 

  Design an AI prompt playgroundAI systemDesign a web workspace where developers write prompts, run them against several models and settings, compare outputs side by side, and save and share versions.

  Last asked Oct 2026Were you asked this too?
• 

  One-on-one chat system with presence and offline queueSystem designDesign direct messaging for single-device users: delivery, presence, offline-recipient queueing, session storage, and the choice of message bus. Group chat is explicitly out of…

  Last asked Oct 2026Were you asked this too?
• 

  Batched LLM inference API over GPUsAI systemDesign an inference API that batches requests across GPUs efficiently. The discussion goes deep on a few axes: utilization, scaling under spikes, and separating prompt processing…

  Last asked Oct 2026Were you asked this too?
• 

  Anthropic culture round: build your point of view on AI safetyBehavioralThe culture round asks what you actually think about AI safety and how it shows up in your work. Read the primary sources, form a short point of view with a chosen top risk, a…

  Last asked Sep 2026Were you asked this too?
• 

  Deduplicate files by size then content hashCodingWalk a directory tree and group files with identical content. Compare file sizes first so large files are hashed only when sizes collide, then hash the rest. Be ready to discuss…

  Last asked Sep 2026Were you asked this too?
• 

  Recruiter screen on why AnthropicBehavioralA substantive first screen that skips the resume tour and opens on motivation: familiarity with the lab's published research and policy stances, plus role fit.

  Last asked Sep 2026Were you asked this too?
• 

  Fix a function-cache key function, then make it durableCodingA small memoizing cache for arbitrary callables has a broken key function that mishandles positional versus keyword arguments and defaults. Fix the keying first, then extend the…

  Last asked Sep 2026Were you asked this too?
• 

  Reconstruct call stacks from enter and exit tracesCodingA profiler gives you call stacks sampled over time, outermost frame first. Turn them into function start and end events, then go the other way: replay start/end events into stacks…

  Last asked Sep 2026Were you asked this too?
• 

  Distributed mode and median across worker shardsCodingYou are given a cluster of k worker nodes, numbered 0 to k - 1. Each worker holds its own unsorted slice (shard) of a large integer dataset. A worker can read only its own shard…

  Last asked Sep 2026Were you asked this too?
• 

  Greedy longest-match tokenizerCodingImplement a greedy longest-match tokenizer. You are given a string text and a list dictionary. Each entry has the form "<key:<id". key is a non-empty token string and id is its…

  Last asked Sep 2026Were you asked this too?
• 

  Design a telemetry pipeline that reconciles inconsistent metric namesSystem designA large fleet of client software already in the field sends telemetry to a data infrastructure team. Each record carries a telemetryname saying what it measures (a request…

  Last asked Sep 2026Were you asked this too?
• 

  Matmul performance modeling on a fixed GPUML designEstimate wall-clock time for matrix multiplications of varying shapes on a given accelerator from arithmetic intensity and memory movement, then extend the model to a multi-layer…

  Last asked Sep 2026Were you asked this too?
• 

  Critique a design doc from the team's backlogSystem designThe interviewer hands over a real design doc and asks for a critique: surface risks, propose alternatives, and say what is missing before this could ship.

  Last asked Sep 2026Were you asked this too?
• 

  Data platform with ingestion and governed accessSystem designDesign a system that ingests large-scale data, processes it, and serves stakeholders with different access needs and sensitivities.

  Last asked Sep 2026Were you asked this too?
• 

  Hiring-manager round on impact and collaborationBehavioralA behavioral round on past project impact, work with non-engineering partners, mentorship, and roadmap influence. Distinct from the values-focused culture round.

  Last asked Sep 2026Were you asked this too?
• 

  Multi-step task with an LLM tool-use loopCodingIn a notebook with docs open, solve a multi-step problem by writing code that calls a language model as a building block: design the prompts and the loop that invokes tools and…

  Last asked Sep 2026Were you asked this too?
• 

  Review a series of pull requestsCodingYou get two or three pull requests, each harder than the last, in Python or Java. Read each one, leave comments as you would on a real review, and point out bugs, design problems…

  Last asked Aug 2026Were you asked this too?
• 

  Project deep dive: present a project you ledBehavioralPresent a project you led in about 20 minutes: context, your role, the architecture, three hard trade-offs, metrics, cost and return, and what you would change. Then take 25…

  Last asked Aug 2026Were you asked this too?
• 

  Task manager with TTL expiry and time-travel snapshotsCodingA four-level task manager: per-user task lists first, then TTL-based auto-deletion, then a snapshot read that returns a user's tasks as they stood at any past timestamp. - Level…

  Last asked Aug 2026Were you asked this too?
• 

  Anthropic OA: leveled in-memory record storeCodingA timed, multi-level exercise building an in-memory store of records with fields: create and read first, then scans and filters, then timestamps with expiry, then backup and…

  Last asked Jul 2026Were you asked this too?
• 

  Repair a looping bootloader programCodingA tiny bootloader stores its program as a list of text instructions. Each entry has the form "<op <value", where value is a signed integer. The machine has one accumulator, which…

  Last asked Jul 2026Were you asked this too?
• 

  Banking system with transfers, cashback and account mergesCodingBuild an in-memory banking system in four levels. Each level adds methods, and every earlier method must keep working. Every call takes a timestamp in milliseconds as its first…

  Last asked Jul 2026Were you asked this too?
• 

  Shift grants with promotions that start next entryCodingTrack workers clocking in and out, apply promotions starting from the next office entry, and total granted bonuses across overlapping work sessions. - Level 1: addworker(id…

  Last asked Jul 2026Were you asked this too?
• 

  Cloud storage with user quotas and backupsCodingDesign an in-memory cloud storage service. The interview has four parts. Each part adds methods, and every earlier method must keep working. Implement a class CloudStorage. File…

  Last asked Jul 2026Were you asked this too?
• 

  Running cost of streamed token usageCodingToken usage for model requests arrives as a stream of usage records rather than all at once. Compute what the usage costs, keeping the totals correct as each new record arrives…

  Last asked Jul 2026Were you asked this too?
• 

  Threshold infection spread on a gridCodingA lab tracks a disease across a rectangular field of cells. You are given an m x n grid where 0 is a healthy cell and 1 is an infected cell, and an integer threshold. Each day…

  Last asked Jul 2026Were you asked this too?
• 

  Recipe manager with ingredient searchCodingBuild an in-memory recipe book. Each recipe has a name, an ordered list of ingredients and an ordered list of steps. The work comes in two parts; finish part 1 before you start…

  Last asked Jun 2026Were you asked this too?
• 

  Bank ledger with two-step transfers and mergesCodingModel simple bank accounts with deposits and withdrawals, a transfer flow that holds funds until the recipient accepts, and an account-merge step that must keep working even if a…

  Last asked Jun 2026Were you asked this too?
• 

  Simplified filesystem with paths and traversalCodingImplement a small filesystem with path lookup and directory traversal, extended in the final level with model-related behavior. The harness is strict, so edge cases around paths…

  Last asked May 2026Were you asked this too?
• 

  Stepwise DNS resolver with aliasingCodingBuild a simplified DNS resolver step by step against a fixed test harness: basic name lookup first, then alias layers on top. Later steps assume the earlier ones work. - Level 1:…

  Last asked May 2026Were you asked this too?
• 

  Weighted data batcher with deterministic resumeCodingBuild a sampler that draws weighted batches from a dataset registry and can resume deterministically from a saved offset, including the edge case where the batch size does not…

  Last asked May 2026Were you asked this too?
• 

  Mechanistic interpretability experiment on double descentML designDesign and run a small-model experiment reproducing sample-aspect double descent, explain the phenomenon from theory, and propose a remedy. The deliverable is the experiment plus…

  Last asked Apr 2026Were you asked this too?
• 

  ML configuration system with inheritance and validationML designDesign and partly implement an ML configuration system covering schemas, inheritance and overrides, validation, and reproducibility. python file=solution.py class Field: def…

  Last asked Apr 2026Were you asked this too?
• 

  Debug a GRPO training script and defend the RL mathML designBelow is a small group relative policy optimization (GRPO) trainer in plain Python. The policy is a table of logits per prompt, and the reward is the share of tokens that match…

  Last asked Mar 2026Were you asked this too?
• 

  QKV attention programming with einsum fluencyML designA short programming screen around sequence models and scaled dot-product attention, plus quick ML knowledge questions. Reading equations and expressing the math with einsum-style…

  Last asked Feb 2026Were you asked this too?
• 

  Self-contained ML take-home with live reviewML designA few-hour modeling problem you solve alone on your own machine, followed by a live review: you present your results, code and trade-offs, and the interviewers ask questions…

  Last asked Feb 2026Were you asked this too?
• 

  Design guardrails and fallbacks for a reliable LLM serviceAI systemA product depends on calls to a large language model. Design the guardrails and fallback paths that keep it reliable: what you check on the way in and on the way out, what the…

  Last asked Feb 2026Were you asked this too?
• 

  Thread-safe task queue on a linked listCodingA task queue was written for one thread: a singly linked list with head and tail, where put appends at the tail and get removes from the head. Explain what goes wrong when many…

  Last asked Feb 2026Were you asked this too?
• 

  Min stackCodingDesign a stack that supports push, pop, top and returning the current minimum element, each in constant time. Explain how the minimum stays correct after pops, how you handle…

  Last asked Feb 2026Were you asked this too?
• 

  Prompt design with evaluation scaffoldingML designIn a notebook, design prompts for a defined task and write the surrounding scaffolding that evaluates them. The exercise tests task framing and measurement more than clever…

  Last asked Jan 2026Were you asked this too?
• 

  IPv4 range iterator with reverse and CIDR supportCodingAn IPv4 address such as "172.16.4.9" is four octets, each 0 to 255. Taken together, the octets form one unsigned 32-bit number. Build iterators that walk IPv4 addresses lazily…

  Last asked Jan 2026Were you asked this too?
• 

  Kernel-optimization take-home scored by speedupCodingA two-hour take-home that replaces the coding screen. AI tools are allowed. You get a simulator for a small accelerator and a working but slow kernel; your score is the speedup in…

  Last asked Dec 2025Were you asked this too?
• 

  Design the front end of a desktop AI chat appSystem designDesign the client side of a desktop chat app for an AI assistant: the conversation list, the chat view and the model's replies as they stream in. This is a front-end system design…

  Possibly retiredWere you asked this too?
• 

  Neural network forward and backward pass from scratchCodingBuild a small neural network and a loss function by hand, without an autodiff library. Write the forward pass, then derive and code the backward pass that computes the gradient of…

  Possibly retiredWere you asked this too?
• 

  Implement a trie (prefix tree)CodingImplement a prefix tree over strings: insert a word, check whether a whole word is present, and check whether any stored word starts with a given prefix. Design the classes so…

  Possibly retiredWere you asked this too?
• 

  Build your story bankBehavioralPrepare 8 to 15 real stories, each tagged with the signals it shows (ownership, conflict, failure, ambiguity, influence, impact, mentorship, quality), so that any behavioral…

  Were you asked this too?
• 

  Design a GPU scheduling platformAI systemShare a large GPU cluster among research and production teams: queue training and inference jobs, enforce quotas and priorities, place jobs on topology-aware hardware, and preempt…

  Were you asked this too?
• 

  Design a RAG system over company docsAI systemEmployees should be able to ask questions about internal documents (Drive, Confluence, Slack, GitHub) and get grounded answers with citations.

  Were you asked this too?
• 

  Design a Slack-like chat systemSystem designWorkspaces, channels, threads and direct messages with real-time delivery, unread state, search and history, for large companies.

  Were you asked this too?
• 

  Design an AI coding agent sandboxAI systemAgents run untrusted code for many users. Design isolated, fast-starting sandboxes with file and network controls.

  Were you asked this too?
• 

  Design an LLM inference serving platformAI systemDesign the system that serves several large models to many customers with low latency and high GPU utilization.

  Were you asked this too?
• 

  Design ChatGPTAI systemDesign a consumer chat product backed by a large language model: conversations, streamed replies, history across devices, at 100M weekly users.

  Were you asked this too?
• 

  Design Facebook MessengerSystem designOne-to-one and group messaging for a billion users: real-time delivery, offline sync, read receipts and message history on every device.

  Were you asked this too?
• 

  Design review: an eval gate for model launchesML designYou are reviewing the design of the evaluation gate that decides whether a new model checkpoint can launch. Critique it: what it can miss, how it can mislead, what you would…

  Were you asked this too?
• 

  Design review: checkpointing a large training runAI systemYou are reviewing the checkpointing design for a month-long training run. Critique it: what fails, what it costs, what you would change, and what the doc does not cover. You have…

  Were you asked this too?
• 

  Design review: rolling out a new model versionAI systemYou are reviewing a launch plan written by the serving team. Critique it: rank the risks, propose changes, and say what must be true before you would approve it. You have 45…

  Were you asked this too?
• 

  Design review: usage billing for an LLM APISystem designYou are reviewing a teammate's design doc before it goes to the wider team. Read it, then lead a critique: what is wrong, how bad each problem is, what you would change, and what…

  Were you asked this too?
• 

  Evaluate a new LLM before launchML designYou have a new model checkpoint. Design the evaluation process that decides whether it ships.

  Were you asked this too?
• 

  Group prompts into templates under concurrent ingestionCodingAn API receives prompts from many threads. Most come from a few templates filled with different values: Summarize ticket 48213 for alice@acme.com and Summarize ticket 1007 for…

  Were you asked this too?
• 

  How do you use AI in your work?BehavioralDescribe how AI tools fit into your engineering work: where they help, where they fail, how you verify their output, and how they have changed the way you or your team work.

  Were you asked this too?
• 

  In-memory note-taking system with keyword searchCodingDesign an in-memory note-taking system, SecondBrainSystem. Each note gets a unique id and carries metadata about when it was created and last modified. Callers pass the current…

  Were you asked this too?
• 

  Minimum workers with first-free job assignmentCodingJobs arrive as (jobid, start, end) with half-open times start, end). Workers are numbered 0, 1, 2, ... Part 1. Process jobs by start time (ties by job id). Each job goes to the…

  Were you asked this too?
• 

  Task manager with user quotas and expiring assignmentsCodingBuild an in-memory TaskManager in four levels. Each level adds methods to the same class, and you move to the next level only after the current one passes. Every method takes an…

  Were you asked this too?
• 

  Tell me about a time you disagreed with your teamBehavioralDescribe a real disagreement about a technical decision: what was at stake, what you did, and how it ended.

  Were you asked this too?