| 2 | weighted LRU cachexAI 2Write a weighted LRU cache with eviction policies | LeetCode 146 LRU Cache / 460 LFU CachePython docs: collections, heapq, bisect, itertools, dataclassesLRU/LFU base; OrderedDict eviction loop by total weight |
| 2 | resumable iteratorOpenAI 2Resumable list iterator: pause a large task halfway and resume later, storing only enough progress state | OpenAI Interview Notes (unofficial reconstructed problems)Fluent Python, 2nd EditionOpenAI notes resumable-iterator; Fluent Python iterator protocol chapter |
| 2 | token-bucket rate limiterAnthropic 1, OpenAI 1, xAI 1Implement a token-bucket rate limiter, then make it distributed. | Scaling your API with rate limitersHow we built rate limiting capable of scaling to millions of domainsToken bucket in Redis; distributed sliding-window counting |
| 2 | async batch processorAnthropic 1, Unspecified frontier 1Write an async batch processor over an API with concurrency limits, retries with jitter and error isolation. | Python docs: asyncio — coroutines and tasks, synchronization primitivesTimeouts, retries, and backoff with jitter (Amazon Builders' Library)Advanced asyncio: Solving Real-world Production Problems (PyCon 2019)Semaphore-bounded TaskGroup; jittered retries; graceful async shutdown |
| 2 | concurrency and multithreadingAnthropic 2How would you parallelise this task? (Concurrency and data mutation come up repeatedly across rounds.) | Python docs: concurrent.futures — launching parallel tasksKeynote on Concurrency (PyBay 2017)Executor patterns; when threads vs processes; shared-state rules |
| 2 | refactor messy code with testsOpenAI 2Refactor bad code: here are ~120 lines of working but messy code with passing tests. Improve the architecture without breaking them. What do… | Refactoring: Improving the Design of Existing Code, 2nd EditionWorking Effectively with Legacy CodeRefactoring catalog under passing tests; characterization tests first |
| 2 | profiler samples slowest functionAnthropic 2Given profiler samples from running code, find the slowest part in the code. | Flame Graphspy-spy: sampling profiler for PythonAggregate samples into inclusive/exclusive time; flame graph reading |
| 2 | spreadsheet cell dependenciesOpenAI 2Implement spreadsheet-style cell notation where formulas reference other cells, with updates propagating to dependent cells when a reference… | OpenAI Interview Notes (unofficial reconstructed problems)Python docs: graphlib.TopologicalSorterLeetCode 631 Design Excel Sum Formulaspreadsheet-dependencies; topological recompute; LC 631 propagation |
| 2 | LRU memoize bugfix durabilityAnthropic 2Memoising LRU Cache: Bug Fix and Durability (3 parts / 55 min) | Anthropic Interview Notes (unofficial reconstructed problems)Files are hardlru-cache-bugfix-durability three parts; crash-safe file writes |
| 2 | profiler samples to call eventsAnthropic 2 · 1 staff+Call Stacks from Profiler Samples: A profiler observes a running program by taking periodic *samples*: | Anthropic Interview Notes (unofficial reconstructed problems)Trace Event FormatSamples to begin/end call events; Trace Event Format nesting |
| 2 | time-based KV storeOpenAI 2Time-Based Key-Value Store: every set includes a timestamp; get returns the value active at a given time. | LeetCode 981 Time Based Key-Value Store / 1146 Snapshot ArrayOpenAI Interview Notes (unofficial reconstructed problems)Python docs: collections, heapq, bisect, itertools, dataclassesLC 981; OpenAI notes time-based-kv-store; bisect by timestamp |
| 2 | GPU/token credit ledgerOpenAI 2 · 1 staff+Design a credit tracking service for user token balances | OpenAI Interview Notes (unofficial reconstructed problems)Stripe Billing credits (credit grants)gpu-credits ledger; production credit-grant model with expiry |
| 1 | progressive 4-level class APIAnthropic 190-minute automated online coding: build a class exposing a public API exactly per spec; 4 levels, each unlocked after passing tests and req… | CodeSignal Industry Coding FrameworkCodeSignal Practice: Industry Coding Framework (mock)Open Gauntlet: 4-level progressive coding practiceICF format; level-gated mocks with exact public API |
| 1 | stack samples to tracesAnthropic 1Transform call stacks or samples into execution traces. | Anthropic Interview Notes (unofficial reconstructed problems)LeetCode 636 Exclusive Time of Functionsstack-trace-reconstruction; stack diffing between samples |
| 1 | LRU cache layeredAnthropic 1Implement an LRU cache. | LeetCode 146 LRU Cache / 460 LFU CacheLC 146 hash map plus doubly linked list |
| 1 | LRU cache file persistenceAnthropic 1How would you handle files? | Files are hardAnthropic Interview Notes (unofficial reconstructed problems)Temp file, fsync, rename; durability part of LRU notes |
| 1 | LRU cache serializationAnthropic 1What are the serialization tradeoffs? | Python docs: pickle, shelve and json (data persistence)Protocol Buffers: Encodingpickle vs JSON vs custom bytes: safety, versioning, speed |
| 1 | image conversion projectAnthropic 1 · 1 staff+Practical coding round: work inside a full project containing image files and implement image conversion, finding and learning an image libr… | Pillow (PIL fork) documentationAnthropic Interview Notes (unofficial reconstructed problems)Pillow conversion APIs; image-processing-pipeline inside a project |
| 1 | in-memory database progressiveAnthropic 1Database-style CodeSignal rounds testing raw implementation speed | Open Gauntlet: 4-level progressive coding practiceLibreSignal: CodeSignal ICF practice frameworkTimed database-style 4-level mocks for raw speed |
| 1 | SQL and pandas data analysisAnthropic 1SQL and Python data work: pull data, clean it, and analyze it | SQLBolt: interactive SQL lessons10 minutes to pandasPython for Data Analysis, 3E (open access)SQL joins/aggregates; pandas cleaning and groupby |
| 1 | data transformationAnthropic 1Final round: Python data-transformation task | Python for Data Analysis, 3E (open access)Regular Expression HOWTO (Python docs) + csv moduleData transformation with pandas or stdlib regex/csv |
| 1 | debug prewritten pipelineAnthropic 1Final round: debugging exercise on a prewritten data pipeline | The Debugging BookPython docs: pdb — the Python debuggerHypothesis-driven debugging, tracing, post-mortem pdb |
| 1 | spreadsheet formulas cache invalidationOpenAI 1Coding: practical Excel/spreadsheet problem around formulas, caching and invalidation | LeetCode 631 Design Excel Sum Formula500 Lines or Less: Contingent — A Fully Dynamic Build SystemBuild Systems à la Carte: Theory and PracticeFormula caching with dirty-marking; build-system recompute strategies |
| 1 | credits systemOpenAI 1Phone screen coding problem around credits | OpenAI Interview Notes (unofficial reconstructed problems)OpenAI notes gpu-credits phone-screen variant |
| 1 | refactoringOpenAI 1Onsite refactoring exercise | Refactoring: Improving the Design of Existing Code, 2nd EditionBeyond PEP 8: Best practices for beautiful intelligible code (PyCon 2015)Named refactorings; what reviewers notice in a live refactor |
| 1 | memory allocatorOpenAI 1Implement a memory allocator (OS-style), defining all data structures yourself | CS:APP3e student labs (Malloc Lab, Proxy Lab)Operating Systems: Three Easy Pieces (v1.10)Malloc lab free lists and coalescing; OSTEP free-space management |
| 1 | class designxAI 1Onsite: practical class-design coding problems rather than LeetCode (one coding round plus a hybrid coding/system-design round) | A Philosophy of Software Design, 2nd EditionPython's Class Development Toolkit (PyCon 2013)awesome-low-level-designDeep modules; class evolving with requirements; OOD problem bank |
| 1 | LRU cachexAI 1Implement an LRU cache | LeetCode 146 LRU Cache / 460 LFU CacheLC 146 LRU cache |
| 1 | read unfamiliar LLM codebasexAI 1Read through and reason about a large LLM-related codebase the interviewer shares (no clarification of inputs/outputs given) | nanoGPT (code-reading practice)vLLM Architecture OverviewTimed read of a small GPT codebase; inference-engine architecture map |
| 1 | ship product with AIxAI 14-hour take-home that is like shipping a quick product with AI | Best practices for Claude CodeBuilding effective agentsAgentic workflow for fast product builds; simplest-working-design patterns |
| 1 | extend unfamiliar classxAI 11-hour in-person coding: understand and extend a huge unfamiliar class quickly | Working Effectively with Legacy CodeRefactoring: Improving the Design of Existing Code, 2nd EditionSeams and characterization tests to extend a large class safely |
| 1 | resumable iterator TDDOpenAI 1Define a templatized iterator supporting suspension and resumption, using TDD (write test cases first, then implementation) | Canon TDDOpenAI Interview Notes (unofficial reconstructed problems)TDD loop; resumable-iterator suspension and resumption |
| 1 | in-memory DB with query opsOpenAI 1implement a DB. add where clause, order by, delete command (via tombstone, LSM) etc | LeetCode 2408 Design SQLAnthropic Interview Notes (unofficial reconstructed problems)LevelDB implementation notesWHERE/ORDER BY on LC 2408; tombstones and LSM from LevelDB notes |
| 1 | balance at timestampOpenAI 1Given an unsorted series of post/delete statements each with a timestamp, compute the balance at any arbitrary timestamp (similar to Time Ba… | LeetCode 981 Time Based Key-Value Store / 1146 Snapshot ArrayPython docs: collections, heapq, bisect, itertools, dataclassesSort events by timestamp; bisect over prefix balances |
| 1 | expiring credit balanceOpenAI 1Implement transaction(ts, id, amount, expiration_timestamp), subtract(ts, amount), get_balance(ts); transactions arrive unordered and remain… | OpenAI Interview Notes (unofficial reconstructed problems)Python docs: collections, heapq, bisect, itertools, dataclassesgpu-credits expiry; heap ordered by expiration |
| 1 | type system for toy languageOpenAI 1Design a type system for a hobby programming language | Type inferenceOpenAI Interview Notes (unofficial reconstructed problems)Constraint generation and unification; OpenAI notes type-inference |
| 1 | multi-level OA system buildAnthropic 190-minute incremental multi-level CodeSignal challenge building one core application/data structure from scratch, progressing from basic fun… | Open Gauntlet: 4-level progressive coding practiceCodeSignal Practice: Industry Coding Framework (mock)Incremental 90-minute multi-level practice |
| 1 | web crawler concurrencyAnthropic 1Web crawler (LeetCode Web Crawler) with BFS; follow-up on multithreading using thread pool executors and asyncio | LeetCode 1236 Web Crawler / 1242 Web Crawler MultithreadedPython docs: concurrent.futures — launching parallel tasksLC 1236 BFS then ThreadPoolExecutor follow-up |
| 1 | image directory processing parallelismAnthropic 1Given a set of images in a directory, run some processing, installing a library of your choice; follow-up on parallelism for many files and … | Python docs: multiprocessing and shared_memoryPillow (PIL fork) documentationProcess pool over an image directory; Pillow processing |
| 1 | stack traceAnthropic 1Phone interview: the Stack Trace problem | Anthropic Interview Notes (unofficial reconstructed problems)stack-trace-reconstruction (the 'Stack Trace' phone screen) |
| 1 | size-bounded LRU cachexAI 1Sized LRU cache: cache has a total size limit and each item has its own size; evict LRU items until the new item fits | LeetCode 146 LRU Cache / 460 LFU CachePython docs: collections, heapq, bisect, itertools, dataclassesSize-bounded eviction on OrderedDict; LRU base |
| 1 | nested list iteratorxAI 1asked about creating iterators on a list and nested lists | LeetCode 341 Flatten Nested List Iterator / 284 Peeking IteratorLC 341 lazy nested-list flattening |
| 1 | LLM token parsingxAI 1parsing tokens for LLM | Hugging Face LLM Course: WordPiece tokenizationAnthropic Interview Notes (unofficial reconstructed problems)Greedy longest-match LLM tokenization; tokenizer notes |
| 1 | typeahead debounce highlightxAI 1TypeAhead with debounce and highlight filtered text | GreatFrontEnd: DebounceReact: Quick Start (react.dev/learn)Debounce implementation; React list filtering and highlight |
| 1 | tetrisxAI 1Assignment that was 'something about tetris' (described as easy, straightforward) | awesome-low-level-designExercism Python trackOOP board-game modelling; small timed simulation exercises |
| 1 | web crawler BFS then concurrencyAnthropic 1Implement a web crawler: interviewer required a single-threaded BFS version first, then kept asking follow-ups on it (prep materials said co… | 500 Lines or Less: A Web Crawler With asyncio CoroutinesLeetCode 1236 Web Crawler / 1242 Web Crawler MultithreadedSingle-threaded BFS first, then concurrency, as in 500L crawler |
| 1 | banking system OAAnthropic 1CodeSignal OA: bank system (multi-level) | Anthropic Interview Notes (unofficial reconstructed problems)Anthropic notes oa-bank-system |
| 1 | image processing parallelismAnthropic 1Batch-process cat images with Pillow (transform call syntax given); part 2: optimize running time — candidate used multiprocessing, asked wh… | Pillow (PIL fork) documentationPython docs: multiprocessing and shared_memoryPython Concurrency From the Ground Up: LIVE! (PyCon 2015)Pillow transforms; CPU-bound work moved to processes |
| 1 | function LRU cache crash recoveryAnthropic 1Coding: LRU cache in two parts — generate a cache key for a function and its arguments; then recover the cache from a crash by writing a wri… | Anthropic Interview Notes (unofficial reconstructed problems)Files are hardAtomic Commit In SQLiteCache-key generation; crash-recovery via journal and atomic rename |
| 1 | LRU cache args + persistenceAnthropic 1Given an implemented Python LRU cache that mishandles *args and **kwargs, fix it; follow-up: make the LRU cache persistent (serialization, e… | CPython source: Lib/functools.py (lru_cache, _make_key)Anthropic Interview Notes (unofficial reconstructed problems)Fix *args/**kwargs keys like CPython; then persistence follow-up |
| 1 | web crawler fragment + thread poolAnthropic 1Web crawler restricted to a single hostname: single-threaded first, then multi-threaded follow-up; must strip URL fragments (#...) before de… | Anthropic Interview Notes (unofficial reconstructed problems)Python docs: urllib.parse (urljoin, urldefrag, urlsplit)concurrent-web-crawler hostname and fragment rules; urldefrag/urljoin |
| 1 | same-host web crawlerAnthropic 1Write a web crawler that only scrapes pages under the same hostname (real website URL given); single-threaded first, then how to make it mul… | LeetCode 1236 Web Crawler / 1242 Web Crawler MultithreadedPython docs: urllib.parse (urljoin, urldefrag, urlsplit)Same-hostname restriction; correct URL parsing |
| 1 | stack trace eventsAnthropic 1Stack Trace problem with many follow-ups | Anthropic Interview Notes (unofficial reconstructed problems)Trace Event Formatstack-trace-reconstruction follow-ups; event nesting invariants |
| 1 | banking system levelsAnthropic 1OA in-memory banking system, 4 levels: add account/pay/deposit; top_activity returning k users by activity (absolute amounts); transfer and … | Anthropic Interview Notes (unofficial reconstructed problems)LeetCode 2043 Simple Bank Systemoa-bank-system levels incl. top_activity ranking |
| 1 | recipe manager OAAnthropic 1OA: recipe manager | Anthropic Interview Notes (unofficial reconstructed problems)Anthropic notes oa-recipe-manager |
| 1 | lru_cache key + crash recoveryAnthropic 1LRU cache used like Python's functools.lru_cache: implement generate_key(*args, **kwargs) to build a hashable key; follow-up: restore the ca… | CPython source: Lib/functools.py (lru_cache, _make_key)Files are hardgenerate_key like lru_cache; crash-safe persistence |
| 1 | PIL image processing parallelAnthropic 1Image processing with PIL (grayscale, scale, resize): small images first, then large images; improve performance with ProcessPoolExecutor | Pillow (PIL fork) documentationPython docs: multiprocessing and shared_memoryPIL grayscale/resize; parallelize large images with processes |
| 1 | image processingAnthropic 1Image processing coding | Pillow (PIL fork) documentationPillow image operations |
| 1 | PR review with Claude CodeAnthropic 1In an online VS Code + GitHub-like site with four pull requests, using Claude Code: explain the problem each PR solves and comment on each; … | Anthropic Interview Notes (unofficial reconstructed problems)Best practices for Claude CodeGoogle Engineering Practices: Code Reviewagentic-pr-review; Claude Code verify workflow; review checklist |
| 1 | string parsing + introspectionCohere 1OA: basic Python string parsing; last question passes a callable and you must print the function's name, its input/output format, and the ex… | Python docs: inspect — signatures and source of live objectsinspect.signature and getsource on a passed callable |
| 1 | GPU usage per dayMistral 1Given GPU clusters as an array (e.g. [10, 20]) and a list of Usage(day, cluster, gpu) objects, output a 2D array of GPUs left per cluster ea… | Python docs: collections, heapq, bisect, itertools, dataclassesEffective Python: 125 Specific Ways to Write Better Python (3rd ed.)Counter/defaultdict aggregation per day and cluster |
| 1 | IPv4 range to CIDROpenAI 1IP address problem in parts; third part converts IPv4 ranges to CIDR blocks; write your own unit tests | OpenAI Interview Notes (unofficial reconstructed problems)LeetCode 751 IP to CIDRPython docs: ipaddress (summarize_address_range, ip_network)ipv4-cidr-iterator; LC 751; summarize_address_range as oracle |
| 1 | structured data stream parsingOpenAI 1Parse and process a structured data stream: basic version, then edge cases, complexity, huge data volume / streaming | HTML Standard: Server-sent events (event stream parsing)Fluent Python, 2nd EditionIncremental line parsing across chunks; generator-based stream processing |
| 1 | full-stack live buildOpenAI 1Build a simple full-stack tool from scratch: React frontend displaying data, backend API, state handling and some real-time updating | React: Quick Start (react.dev/learn)FastAPI Tutorial — User GuideSmall React frontend plus FastAPI backend |
| 1 | monster battleOpenAI 1"Monster battle" coding question | OpenAI Interview Notes (unofficial reconstructed problems)OpenAI notes monster-battle |
| 1 | Django rate limiter debugOpenAI 1Debugging session: find what is wrong with the rate limiter in a Django app | OpenAI Interview Notes (unofficial reconstructed problems)Scaling your API with rate limitersrate-limiter-debug; production token-bucket semantics |
| 1 | grid infection simulationOpenAI 1Simulation: n x m grid where a healthy person becomes infected if they have N infected neighbors; how long until everyone is infected; five … | OpenAI Interview Notes (unofficial reconstructed problems)LeetCode 994 Rotting Oranges / 289 Game of LifeN-infected-neighbour threshold rule; LC 289 synchronous updates |
| 1 | concurrent job scheduler debugOpenAI 1Debug a buggy Python job scheduler: data races, deadlocks, lock contention, rate limiter correctness; write tests, then compute scheduling t… | Operating Systems: Three Easy Pieces (v1.10)Python docs: threading and queueThe Debugging BookConcurrency-bug chapter; locks/conditions; systematic debugging |
| 1 | plant infection with burningOpenAI 1Plant infection grid: part 4 adds dead countdown for >=K infected neighbors and recovery countdown for >=T and <K; part 5 choose rows/column… | OpenAI Interview Notes (unofficial reconstructed problems)infection-spread part with death countdowns and recovery |
| 1 | infection spread with immunityOpenAI 1General coding (CoderPad, stdlib only): 5-part grid infection spread simulation — part 2 adds immune cells, part 3 adds a days-to-recovery c… | OpenAI Interview Notes (unofficial reconstructed problems)infection-spread with immunity variants |
| 1 | extensible card hand rankingOpenAI 1Card game: compare hands (three of a kind > pair > high card); add flush (three of a kind > flush > pair > high card); add a discard rule (d… | OpenAI Interview Notes (unofficial reconstructed problems)OpenAI notes poker-hands extensible ranking |
| 1 | plant infection countdownsOpenAI 160-minute plant-infection coding problem in parts; part 3 computes countdowns for recovery and immunity (initially infected plants also reco… | OpenAI Interview Notes (unofficial reconstructed problems)infection-spread countdown parts |
| 1 | rectangle board with undoOpenAI 1Rectangle: maintain a 2D board with add rectangle, remove rectangle, undo, then follow-ups like removeAt and batch remove (multipart, state … | Refactoring.Guru: Design Patterns in Pythonawesome-low-level-designCommand pattern for undo; board modelling |
| 1 | expiring GPU credit ledgerOpenAI 1GPU credit problem: add_credit, subtract, get_balance; subtract uses earliest-expiring credits first; 4 given test cases, then optimization … | OpenAI Interview Notes (unofficial reconstructed problems)Python docs: collections, heapq, bisect, itertools, dataclassesgpu-credits earliest-expiry consumption; heap by expiration |
| 1 | in-memory KV store serializationOpenAI 1KV store coding problem (recruiter said in-memory data store; know how to serialize it in Python) | Python docs: pickle, shelve and json (data persistence)Python docs: struct — interpret bytes as packed binary dataIn-memory KV serialization choices; struct packing |
| 1 | KV storeOpenAI 1KV store coding | OpenAI Interview Notes (unofficial reconstructed problems)Let's Build a Simple Databasein-memory-database notes; simple DB structure |
| 1 | job scheduler debuggingOpenAI 1Debug a Python job scheduler repo from bug reports, writing a test for each fix: failed jobs stuck in PENDING because of missing exception h… | The Debugging Bookpytest documentationWorking Effectively with Legacy CodeRepro test per bug report, then fix; characterization tests |
| 1 | distributed node counting via messagesOpenAI 1Given a root node and sendAsyncMessage(nodeId, message), implement receiveMessage(fromNodeId, message) to count all nodes in the network tre… | OpenAI Interview Notes (unofficial reconstructed problems)OpenAI notes cluster-count-topology message-passing count |
| 1 | chatbot refactorOpenAI 1Refactor a chatbot (coding); variant: for each away user, when someone @s them the bot replies with their away status | OpenAI Interview Notes (unofficial reconstructed problems)OpenAI notes chat-bot-refactoring |
| 1 | malloc/free logarithmicOpenAI 1Implement malloc/free with provided tests; part 2: optimize to O(log n) malloc and free (e.g. two sorted lists, third-party library allowed) | Sorted Containers (SortedList, SortedDict)OpenAI Interview Notes (unofficial reconstructed problems)LeetCode 2502 Design Memory AllocatorSorted free lists by address and size for O(log n) |
| 1 | fault-tolerant queue with DLQOpenAI 1Fault-tolerant queue: implement enqueue, reserve, fail and a dead-letter queue passing given unit tests; part 2 adds dependencies between ta… | OpenAI Interview Notes (unofficial reconstructed problems)Amazon SQS: visibility timeout and dead-letter queuesHow to do distributed lockingfault-tolerant-work-queue; visibility timeout and DLQ semantics |
| 1 | fault-tolerant work queueOpenAI 1Fault-tolerant work queue (two parts) | OpenAI Interview Notes (unofficial reconstructed problems)Amazon SQS: visibility timeout and dead-letter queuesfault-tolerant-work-queue two parts; redrive policy |
| 1 | time-windowed GPU creditsOpenAI 1GPU credit system: grant credits with activation and expiration times, consume credits at a time, get balance at a time; operations may arri… | OpenAI Interview Notes (unofficial reconstructed problems)Stripe Billing credits (credit grants)gpu-credits activation/expiry windows; credit grants model |
| 1 | grid infection with deathsOpenAI 1Classic infection problem (multi-part; part 4 returns number of days and final number of deaths) | OpenAI Interview Notes (unofficial reconstructed problems)infection-spread part 4: days and death counts |
| 1 | fault-tolerant queue cooldownsOpenAI 1Fault-tolerant queue: enqueue, reserve, fail, DLQ with 7 unit tests; part 2 adds task dependencies; preempted tasks have a cooldown computed… | OpenAI Interview Notes (unofficial reconstructed problems)Python docs: collections, heapq, bisect, itertools, dataclassesfault-tolerant-work-queue cooldowns; heap ordered by enqueue time |
| 1 | contagion grid simulationOpenAI 1Contagion: part 1 time until everyone infected; part 2 add immune state; part 3 become immune D days after infection, time until stable; par… | OpenAI Interview Notes (unofficial reconstructed problems)LeetCode 994 Rotting Oranges / 289 Game of LifeContagion time-to-all, immunity; BFS by day |
| 1 | friend graph snapshotsOpenAI 1Friend Circle: follow/unfollow/snapshot; return all friends at a snapshot; recommend friends from a snapshot; diff friends between two snaps… | OpenAI Interview Notes (unofficial reconstructed problems)LeetCode 981 Time Based Key-Value Store / 1146 Snapshot Arraysocial-network snapshots and recommendations; snapshot array |
| 1 | social media feedOpenAI 1Design a social media feed (coding, parts 1 and 2) | LeetCode 355 Design TwitterLC 355 follow graph and merged feed |
| 1 | build a TerraformOpenAI 1Extra round: "Building a Terraform" problem | Python docs: graphlib.TopologicalSorter500 Lines or Less: Contingent — A Fully Dynamic Build SystemBuild Systems à la Carte: Theory and PracticeDependency DAG apply order; dynamic rebuild of changed resources |
| 1 | snapshot setOpenAI 1Implement a Set that supports snapshots: add/remove, create a snapshot, read the set state at a snapshot | LeetCode 981 Time Based Key-Value Store / 1146 Snapshot ArrayOpenAI Interview Notes (unofficial reconstructed problems)LC 1146 snapshot ids; social-network snapshot set |
| 1 | IP address / CIDROpenAI 1IP address coding problem (5 parts, including an IPv6 part) | Python docs: ipaddress (summarize_address_range, ip_network)OpenAI Interview Notes (unofficial reconstructed problems)IPv4 and IPv6 networks; ipv4-cidr-iterator parts |
| 1 | sliding-window chat analyticsOpenAI 1Chat session analytics over a non-decreasing log stream: ingest(user, chat, ts) and count_recent(user, chat) for the closed window [T-14, T]… | OpenAI Interview Notes (unofficial reconstructed problems)LeetCode 359 Logger Rate Limiter / 362 Design Hit Counterevent-window-aggregation; hit-counter sliding window |
| 1 | social graph snapshotsOpenAI 1Multipart social graph + snapshot problem: first part as on Hack2Hire, then getFollowers + getFollowees, then a recommendation part | OpenAI Interview Notes (unofficial reconstructed problems)OpenAI notes social-network multipart snapshots |
| 1 | GPU credit overdraftOpenAI 1GPU credit II: spend earliest-expiring credit first; if the balance ever goes negative, all later get_balance calls return None; follow-up: … | OpenAI Interview Notes (unofficial reconstructed problems)Accounting for Developers, Part Igpu-credits overdraft; ledger entries instead of mutable balance |
| 1 | counting machinesOpenAI 1Complex challenge around counting machines | OpenAI Interview Notes (unofficial reconstructed problems)cluster-count-topology machine counting |
| 1 | processes vs threadsxAI 1What's the difference between multiprocessing and multithreading? | Python behind the scenes #13: the GIL and its effects on Python multithreadingPython docs: multiprocessing and shared_memoryPython support for free threading (HOWTO) + PEP 703GIL mechanics; process model; free-threaded builds |
| 1 | rocket simulationxAI 1Simulation of moving a rocket in a game-like setting, implemented under a tight time box | Advent of Code (2020 Day 8: Handheld Halting)Effective Python: 125 Specific Ways to Write Better Python (3rd ed.)Time-stepped simulation drills; clean class structure |
| 1 | race condition depositxAI 1Debug a multithreaded bank account deposit program (ThreadPoolExecutor depositing 500 and 700): find the read-sleep-write race condition and… | Python docs: threading and queuePython Concurrency From the Ground Up: LIVE! (PyCon 2015)LeetCode Concurrency problem set (e.g. 1114 Print in Order, 1188 Bounded Blocking Queue)Lock around read-modify-write; why deposits race |
| 1 | three-stage problem with concurrencyAnthropic 1Technical assessment (after CodeSignal): a popular LeetCode-style question set up as three stages, including concurrency | Python docs: concurrent.futures — launching parallel tasksLeetCode Concurrency problem set (e.g. 1114 Print in Order, 1188 Bounded Blocking Queue)Executor-based staging; synchronization puzzles |
| 1 | single-file refactorOpenAI 1 · 1 staff+Onsite code-refactoring round: refactor a single file in CoderPad | Refactoring: Improving the Design of Existing Code, 2nd EditionBeyond PEP 8: Best practices for beautiful intelligible code (PyCon 2015)Single-file refactor under tests |
| 1 | progressive CodeSignal levelsAnthropic 1 · 1 staff+Automated CodeSignal screen (Python only): progressive levels where only passing all visible unit tests matters; perfect score required | CodeSignal Industry Coding FrameworkOpen Gauntlet: 4-level progressive coding practiceAll-tests-pass level gating; timed mocks |
| 1 | no built-ins constraintAnthropic 1Live coding on a CodeSignal-style problem with the constraint 'do not use any inbuilt methods' (e.g. implementing string splitting yourself) | Exercism Python trackMIT 6.101 Fundamentals of Programming (Fall 2026)Implement split, parsing helpers without built-ins |
| 1 | extend pre-written codeOpenAI 1Additional coding round after onsite: work on pre-written code with two interviewers | Working Effectively with Legacy CodeRefactoring: Improving the Design of Existing Code, 2nd EditionWork inside pre-written code safely |
| 1 | debug and improve codeOpenAI 1Coding round: a debugging / improvement exercise on existing code | The Debugging BookGoogle Engineering Practices: Code ReviewDebug then improve; review checklist |
| 1 | pandas data codingOpenAI 1Initial screen: pandas-style data coding | 10 minutes to pandasPython for Data Analysis, 3E (open access)pandas selection, groupby, cleaning |
| 1 | extend class follow-upsxAI 1Technical rounds: implement a class, then extend it through follow-ups | A Philosophy of Software Design, 2nd EditionOpen Gauntlet: 4-level progressive coding practiceInterfaces that absorb follow-ups; progressive practice |
| 1 | progressive Python project with testsAnthropic 1Anthropic Fellows OA: progressive 90-minute Python coding project with unittest-based tests and possible concurrency/threading components | Python docs: unittestOpen Gauntlet: 4-level progressive coding practiceunittest-based level tests; 90-minute progressive mocks |
| 1 | core Python live codingAnthropic 1Anthropic Safety Fellows: 55-minute live coding challenge using only core Python (no ML libraries) | Effective Python: 125 Specific Ways to Write Better Python (3rd ed.)Practical Python ProgrammingCore-Python idioms without libraries |
| 1 | in-memory system merges and historyAnthropic 1Progressive 4-level CodeSignal exercise building an in-memory banking-style system in standard-library Python; the final level required stat… | Anthropic Interview Notes (unofficial reconstructed problems)Open Gauntlet: 4-level progressive coding practiceoa-bank-system merges and history levels |
| 1 | concurrency follow-upAnthropic 1Web crawler phone-screen problem (shown as an image; commenters identify it as crawling linked URLs) with a follow-up to add concurrency; ca… | LeetCode 1236 Web Crawler / 1242 Web Crawler MultithreadedPython docs: concurrent.futures — launching parallel tasksCrawler concurrency follow-up with executors |
| 1 | code review and debuggingAnthropic 1Tech phone screen: first 30 minutes code review and debugging of existing code, second 30 minutes a non-LeetCode coding problem | Google Engineering Practices: Code ReviewThe Debugging BookCode review then debugging of existing code |
| 1 | progressive implement and refactorAnthropic 1CodeSignal OA: 4 modules testing how fast you can implement and refactor code on a simple problem | Open Gauntlet: 4-level progressive coding practiceLibreSignal: CodeSignal ICF practice frameworkImplement-and-refactor 4-level mocks |
| 1 | debugging and time-based trackingOpenAI 1Phone screen topics given by recruiter for a backend role: debugging a production application; debugging a high-scale concurrent distributed… | The Debugging BookLeetCode 981 Time Based Key-Value Store / 1146 Snapshot ArrayProduction debugging method; time-indexed tracking |
| 1 | in-memory data storeOpenAI 1Implement an in-memory data store class with additional functions added in follow-up parts; instructions said performance did not matter | Anthropic Interview Notes (unofficial reconstructed problems)oa-in-memory-database data store with follow-ups |
| 1 | in-memory NoSQL databaseAnthropic 1Code an in-memory NoSQL database (CodeSignal-style progressive task) | Anthropic Interview Notes (unofficial reconstructed problems)LibreSignal: CodeSignal ICF practice frameworkNoSQL-style in-memory DB levels |
| 1 | stack trace to call intervalsAnthropic 1Turn an ordered list of stack-trace strings into start/end points for each function call (easy-medium stack problem), followed by a hard fol… | LeetCode 636 Exclusive Time of FunctionsAnthropic Interview Notes (unofficial reconstructed problems)Stack strings to start/end intervals |
| 1 | image pipeline from JSON specAnthropic 1Transform images based on JSON data in a pipeline using the Pillow library (question picked based on candidate's resume) | Anthropic Interview Notes (unofficial reconstructed problems)Pillow (PIL fork) documentationJSON-recipe image pipeline with Pillow |
| 1 | progressive CRUD app refactorAnthropic 190-minute OA: 4-level CRUD-style app with test cases per level; later levels require refactoring earlier code while adding features | Open Gauntlet: 4-level progressive coding practiceCodeSignal Practice: Industry Coding Framework (mock)CRUD levels forcing refactor |
| 1 | CRUD with TTL and quotaAnthropic 1OA progressive problem combining CRUD + TTL + quota; level 3 needed a helper function used across all answers | Anthropic Interview Notes (unofficial reconstructed problems)LeetCode 1797 Design Authentication Manageroa-cloud-storage quotas; TTL expiry |
| 1 | web scraper with concurrencyAnthropic 1Build a web scraper for a given domain, first serially, then add concurrency/parallelization | 500 Lines or Less: A Web Crawler With asyncio CoroutinesPython docs: asyncio — coroutines and tasks, synchronization primitivesSerial then concurrent scraper with bounded tasks |
| 1 | class design implementationOpenAI 1First-round coding: a class design question, not algorithmically hard but requiring a lot of implementation | A Philosophy of Software Design, 2nd Editionawesome-low-level-designClean class design; OOD practice |
| 1 | routing problem levelsAnthropic 1Anthropic Fellows OA1: a multi-level "routing" problem on CodeSignal | Open Gauntlet: 4-level progressive coding practiceLibreSignal: CodeSignal ICF practice frameworkMulti-level OA format practice (routing variant not public) |
| 1 | incremental live codingAnthropic 1Incremental (multi-part) live coding problem, 55 minutes with an interviewer, explicitly not LeetCode-style | Anthropic Interview Notes (unofficial reconstructed problems)Open Gauntlet: 4-level progressive coding practiceMulti-part live problems; incremental extension |
| 1 | multi-file bug fixing with AIMeta (org unspecified) 1"Coding with AI" round (1 hour): given a 4–5 file software project with test cases and an AI agent, fix bugs distributed across files to pas… | Best practices for Claude CodeThe Debugging BookAgent-assisted multi-file fix with tests as oracle |
| 1 | KV state serializationOpenAI 1Given key/value pairs and helper methods that convert strings to and from bytes, efficiently store that state in the cloud and retrieve it; … | Python docs: struct — interpret bytes as packed binary dataProtocol Buffers: EncodingOpenAI Interview Notes (unofficial reconstructed problems)Bytes helpers, framing, efficient sync of KV state |
| 1 | spreadsheet dependency propagationOpenAI 1Implement Excel-like formulas where updates propagate to dependent cells; how to handle cell dependencies and what to cache and when to inva… | OpenAI Interview Notes (unofficial reconstructed problems)Python docs: graphlib.TopologicalSorterLeetCode 631 Design Excel Sum Formulaspreadsheet-dependencies propagation and cycle detection |
| 1 | tokenizer without delimitersAnthropic 1Write a tokenizer and a detokenizer for a language with no delimiters (tokens like 'a', 'b', 'cd'; input 'abcdaa'); then handle overlapping … | Anthropic Interview Notes (unofficial reconstructed problems)LeetCode 208 Implement Trie / 139 Word Breaklongest-match-tokenizer; overlapping tokens via Word Break DP |
| 1 | banking query processorAnthropic 1Process queries CREATE_ACCOUNT, DEPOSIT and TRANSFER with timestamps and return string results ("true", balances, or "" for invalid ops such… | Anthropic Interview Notes (unofficial reconstructed problems)LeetCode 2043 Simple Bank SystemCREATE/DEPOSIT/TRANSFER query processing levels |
| 1 | exclusive time / duplicate filesAnthropic 1Coding questions like exclusive time of functions and duplicate file finder | LeetCode 636 Exclusive Time of FunctionsLeetCode 609 Find Duplicate File in SystemLC 636 and LC 609 |
| 1 | in-memory DB with TTLAnthropic 1OA: in-memory database in 4 levels — get/set/delete; search and filter; TTL auto-expiration; time-travel queries for historical data | Anthropic Interview Notes (unofficial reconstructed problems)LeetCode 1797 Design Authentication Manageroa-in-memory-database TTL level |
| 1 | cloud storage systemAnthropic 1OA: cloud storage system in 4 stages — add/copy/read files; pattern-matching file search; user management and storage limits; compress and r… | Anthropic Interview Notes (unofficial reconstructed problems)LeetCode 588 Design In-Memory File System / 71 Simplify Pathoa-cloud-storage; LC 588 file system trie |
| 1 | banking systemAnthropic 1OA: banking system in 4 levels — create account/deposit/transfer; rank by spending; scheduled payments and cashback; merge accounts and hist… | Anthropic Interview Notes (unofficial reconstructed problems)oa-bank-system four levels |
| 1 | employee trackerAnthropic 1OA: employee management system — track arrival/departure; hours worked and top performers; future-dated promotions; double-pay periods | Open Gauntlet: 4-level progressive coding practiceawesome-low-level-designProgressive tracker-style systems; OOD modelling |
| 1 | batch image processorAnthropic 1Batch-edit images in a folder according to a JSON config | Anthropic Interview Notes (unofficial reconstructed problems)Pillow (PIL fork) documentationimage-processing-pipeline JSON config |
| 1 | tokenizer bug fixAnthropic 1Given buggy tokenize and detokenize functions, find and fix the bug | Anthropic Interview Notes (unofficial reconstructed problems)The Debugging BookTokenizer bug-fix part; minimal repro |
| 1 | token usage costAnthropic 1Calculate LLM API token costs with different input vs output prices and token counting for streaming responses | Python docs: decimal — fixed-point arithmeticExact cost arithmetic with Decimal |
| 1 | stack samples to trace eventsAnthropic 1Convert call-stack snapshots from a sampling profiler into an event timeline | Anthropic Interview Notes (unofficial reconstructed problems)Trace Event FormatSamples to event timeline |
| 1 | bootloader instruction loopAnthropic 1Fix an instruction loop in a bootloader | Anthropic Interview Notes (unofficial reconstructed problems)Advent of Code (2020 Day 8: Handheld Halting)bootloader-loop-fix; AoC 2020 day 8 loop detection |
| 1 | image pipeline multiprocessingAnthropic 1Image pipeline using PIL for resizing and rotation, then add multiprocessing | Python docs: multiprocessing and shared_memoryPillow (PIL fork) documentationPIL resize/rotate then multiprocessing |
| 1 | longest-match shortcut tokenizerAnthropic 1 · 1 staff+Phone screen: tokenize an input command scanning left to right with predefined shortcut patterns, always preferring the longest matching pat… | LeetCode 208 Implement Trie / 139 Word BreakAnthropic Interview Notes (unofficial reconstructed problems)Trie longest-match scan over shortcuts |
| 1 | JSON-recipe image pipelineAnthropic 1 · 1 staff+Coding: process a collection of images according to transformation recipes defined in JSON, producing an output per image-transformation pai… | Anthropic Interview Notes (unofficial reconstructed problems)image-processing-pipeline recipes |
| 1 | token cost trackingAnthropic 1Implement token-usage cost tracking for input and output tokens, like an LLM API would | Python docs: decimal — fixed-point arithmeticPython docs: collections, heapq, bisect, itertools, dataclassesPer-model input/output pricing aggregation |
| 1 | progressive CodeSignal variantsAnthropic 190-minute 4-level CodeSignal: variants include in-memory DB, banking transactions, student course and grade management, file system, cloud f… | LibreSignal: CodeSignal ICF practice frameworkOpen Gauntlet: 4-level progressive coding practiceAnthropic Interview Notes (unofficial reconstructed problems)Many 4-level variants: DB, bank, storage |
| 1 | Spaces active-time from logsxAI 145-min phone screen: given unsorted Twitter Spaces event logs (create, join, leave), compute total active time per Space; handle unsorted lo… | LeetCode 636 Exclusive Time of FunctionsPython docs: collections, heapq, bisect, itertools, dataclassesSort events; join/leave intervals into active time |
| 1 | KV cache custom serializationxAI 1Build a durable key-value cache with custom byte-level serialization (no JSON/pickle, binary data) | Python docs: struct — interpret bytes as packed binary dataProtocol Buffers: EncodingOpenAI Interview Notes (unofficial reconstructed problems)Custom byte serialization; durable-kv-store |
| 1 | radix tree prefix cachexAI 1Design a radix-tree cache with prefix lookups | SGLang: Efficient Execution of Structured Language Model Programs (RadixAttention)LeetCode 208 Implement Trie / 139 Word BreakRadixAttention prefix tree with LRU; trie base |
| 1 | production concurrency codexAI 1Write production-level concurrency code live instead of a standard LeetCode problem | Python Concurrency From the Ground Up: LIVE! (PyCon 2015)Notes on structured concurrency, or: Go statement considered harmfulPython docs: asyncio — coroutines and tasks, synchronization primitivesProduction-grade structured concurrency |
| 1 | multi-threading follow-upsUnspecified frontier 1Follow-ups after a working solution: how do you support multi-threading? how would you change the solution under a new list of constraints? | Python docs: concurrent.futures — launching parallel tasksKeynote on Concurrency (PyBay 2017)Multithreading follow-ups and shared state |
| 1 | versioned graph snapshotsOpenAI 1Versioned follow graph with snapshot queries: update current relationships while answering historical queries without copying the whole grap… | OpenAI Interview Notes (unofficial reconstructed problems)social-network versioned snapshots |
| 1 | GPU credit ledger out-of-orderOpenAI 1GPU credit ledger: credits expire, usage events may arrive late, query balance at an earlier point in time | OpenAI Interview Notes (unofficial reconstructed problems)Accounting for Developers, Part ILate events; ledger replay by timestamp |
| 1 | durable KV store with WALOpenAI 1Durable string key-value store that survives a crash, including mid-write (atomic writes, write-ahead logging, recovery) | OpenAI Interview Notes (unofficial reconstructed problems)Atomic Commit In SQLiteFiles are harddurable-kv-store WAL; atomic writes |
| 1 | streaming token differOpenAI 1Token-level streaming differ with rollback: text arrives token by token, parts may later be removed or replaced; update only what changed | Python docs: difflib — helpers for computing deltasOpcode diffs as reference for streaming differ |
| 1 | memory allocator free listOpenAI 1Memory allocator: find a large enough free block, split on allocation, merge neighbours on free | Operating Systems: Three Easy Pieces (v1.10)CS:APP3e student labs (Malloc Lab, Proxy Lab)OpenAI Interview Notes (unofficial reconstructed problems)Split, coalesce, free list |
| 1 | GPU credits out-of-orderOpenAI 1Implement GPU credit management: add_credit/subtract/get_balance with [start, expiration) windows, out-of-order timestamps, deduct earliest-… | OpenAI Interview Notes (unofficial reconstructed problems)gpu-credits out-of-order variant |
| 1 | key-value storeOpenAI 1Key-value store implementation | Let's Build a Simple DatabaseOpenAI Interview Notes (unofficial reconstructed problems)KV store implementation basics |
| 1 | GPU credit calculatorOpenAI 1GPU credit calculator: addCredit(creditID, amount, timestamp, expiration), useCredit(timestamp, amount), getBalance(timestamp), with API cal… | OpenAI Interview Notes (unofficial reconstructed problems)Stripe Billing credits (credit grants)addCredit/useCredit with expiration ordering |
| 1 | refactoring and code reviewOpenAI 1Refactoring and code-review style coding questions (common for experienced candidates), plus concurrency problems | Refactoring: Improving the Design of Existing Code, 2nd EditionGoogle Engineering Practices: Code ReviewRefactoring and review rounds |
| 1 | infection thresholds follow-upsOpenAI 1Further parts of the infection-grid problem: infectionThreshold (minimum infected neighbours before a cell is infected) and deathThreshold (… | OpenAI Interview Notes (unofficial reconstructed problems)infection threshold follow-ups |
| 1 | in-memory DB with concurrencyAnthropic 1In-memory database simulation with concurrency handling | Python docs: threading and queueOperating Systems: Three Easy Pieces (v1.10)Locking an in-memory DB; lock granularity |
| 1 | race condition debuggingAnthropic 1Identify and fix race conditions in multithreaded code | Operating Systems: Three Easy Pieces (v1.10)Python docs: threading and queueLeetCode Concurrency problem set (e.g. 1114 Print in Order, 1188 Bounded Blocking Queue)Race identification and fixes |
| 1 | multi-tier bank ledgerAnthropic 1Multi-tier bank ledger system | Accounting for Developers, Part IDesigning robust and predictable APIs with idempotencyAnthropic Interview Notes (unofficial reconstructed problems)Double-entry ledger tiers; bank OA |
| 1 | same-domain web crawlerAnthropic 1Web crawler for same-domain URLs using BFS | LeetCode 1236 Web Crawler / 1242 Web Crawler MultithreadedPython docs: urllib.parse (urljoin, urldefrag, urlsplit)Same-domain BFS crawler |
| 1 | mini database with transactionsAnthropic 1Code a full-featured mini-database with transactions in about 70 minutes | Let's Build a Simple DatabaseOpenAI Interview Notes (unofficial reconstructed problems)500 Lines or Less: DBDB — Dog Bed DatabaseMini DB with transactions and persistence |
| 1 | progressive API implementationAnthropic 1Anthropic Fellows online coding (90 min): automated API implementation challenge with four levels of progressive complexity, speed over opti… | Open Gauntlet: 4-level progressive coding practicePython docs: unittestAutomated API levels with tests |
| 1 | distributed KV storexAI 1Design a distributed key-value store | Dynamo: Amazon's Highly Available Key-value Store (SOSP 2007)Redis persistence (RDB and AOF)Partitioning, replication, persistence trade-offs |
| 1 | multithreaded sortxAI 1Multi-threaded array sort | Python docs: multiprocessing and shared_memoryPython docs: concurrent.futures — launching parallel tasksChunked parallel sort then k-way merge |
| 1 | labeling task scheduler fairnessOpenAI 1Data Labeling Task Scheduler (two parts): build a scheduler that balances data-labeling task assignments across models and human labelers, w… | OpenAI Interview Notes (unofficial reconstructed problems)labeling-task-scheduler fairness |
| 1 | shard rebalancingOpenAI 1Coding: Shard Rebalancing | OpenAI Interview Notes (unofficial reconstructed problems)Dynamo: Amazon's Highly Available Key-value Store (SOSP 2007)shard-rebalance; consistent hashing |
| 1 | banking system 4 levelsAnthropic 1CodeSignal 90-min banking system: L1 create_account and deposit; L2 sort accounts by total transaction volume; L3 transfer and accept_transf… | Anthropic Interview Notes (unofficial reconstructed problems)oa-bank-system four levels |
| 1 | instruction loop detectionAnthropic 1 · 1 staff+Bootloader-repair style problem: simulate a list of instructions and detect where execution starts repeating. | Advent of Code (2020 Day 8: Handheld Halting)Anthropic Interview Notes (unofficial reconstructed problems)Instruction simulation and loop detection |
| 1 | versioned follow graphOpenAI 1 · 1 staff+Follow Graph with Snapshots and Recommendations. | OpenAI Interview Notes (unofficial reconstructed problems)social-network follow graph snapshots |
| 1 | token credit ledgerOpenAI 1 · 1 staff+Credit tracking service for user token balances. | OpenAI Interview Notes (unofficial reconstructed problems)Accounting for Developers, Part IToken credit ledger with entries |
| 1 | streaming SSE chunk parserCohere 1Write a streaming SSE/JSON parser that handles arbitrary chunk boundaries. | HTML Standard: Server-sent events (event stream parsing)Spec line-parsing algorithm across chunk boundaries |
| 1 | task schedulerAnthropic 1Create a task scheduler. | LeetCode 1834 Single-Threaded CPU / 621 Task SchedulerPython docs: collections, heapq, bisect, itertools, dataclassesHeap-driven task scheduling |
| 1 | OOP course/grade managementAnthropic 1Build an OOP system for managing courses, grades and students. | awesome-low-level-designPython's Class Development Toolkit (PyCon 2013)OOP course/grade system modelling |
| 1 | rate limiter evolving requirementsAnthropic 1Build a rate limiter. Every ten minutes I add a requirement: per-tenant limits, burst allowances, then a sliding window. How do you keep the… | Scaling your API with rate limitersHow we built rate limiting capable of scaling to millions of domainsLeetCode 359 Logger Rate Limiter / 362 Design Hit CounterPer-tenant, burst and sliding-window extensions |
| 1 | concurrent LLM calls with 429 retriesAnthropic 1You need to run an LLM call over 50,000 documents. The API allows ~100 concurrent requests and occasionally returns 429s and timeouts. Write… | Python docs: asyncio — coroutines and tasks, synchronization primitivesTimeouts, retries, and backoff with jitter (Amazon Builders' Library)Semaphore at ~100 concurrent; 429 retries with jitter |
| 1 | SQL product pairs bought togetherAnthropic 1SQL: write a query to find the top five pairs of products most frequently purchased together. | SQLBolt: interactive SQL lessonsPostgreSQL tutorial: Window FunctionsSelf-join pairs, group, order, limit |
| 1 | SQL overlapping subscription rangesAnthropic 1SQL: determine whether any user has overlapping subscription date ranges. | SQLBolt: interactive SQL lessonsPostgreSQL tutorial: Window FunctionsSelf-join on overlapping date ranges or LAG |
| 1 | SQL current salary after ETL errorAnthropic 1SQL: return each employee's current salary after an ETL error inserted a new salary row every year. | PostgreSQL tutorial: Window FunctionsROW_NUMBER over latest salary per employee |
| 1 | database ORM step by stepOpenAI 1Create a database ORM, step by step. | Descriptor Guide (Python HOWTO)Architecture Patterns with Python (Cosmic Python)Descriptor-based mini ORM; repository pattern |
| 1 | web crawler in GoOpenAI 1Code a trivial web crawler using Go. | A Tour of Go — Exercise: Web CrawlerGo concurrent crawler exercise |
| 1 | UI from mockupOpenAI 1Implement a UI from a mockup with provided CSS and API. | React: Quick Start (react.dev/learn)React components from mockup |
| 1 | bigram parsingGoogle DeepMind 1Parse bigrams: extract two-word phrases from strings for NLP feature engineering. | Python docs: collections, heapq, bisect, itertools, dataclassesCounter with itertools.pairwise for bigrams |
| 1 | AI-assisted multi-file bug fixMeta MSL 1You're dropped into an unfamiliar multi-file codebase with a failing behaviour and an LLM assistant available. Walk me through how you'd fix… | Best practices for Claude CodeThe Debugging BookAgent-assisted reproduce, fix, verify |
| 1 | stubbed inference scheduler methodxAI 1Here is a scheduler class from a small LLM inference engine. One method, \_admit_requests, is a stub: no spec, no docstring, no tests. Walk … | vLLM Architecture OverviewScheduler and KV-cache admission design in vLLM |
| 1 | four-hour AI product buildxAI 1You have four hours to build and demo a working AI-powered product. How do you spend them? | Best practices for Claude CodeBuilding effective agentsYour AI Product Needs EvalsPlan, build simplest agent, evaluate within four hours |
| 1 | token rate limiter multi-tenantCohere 1Design a token-based rate limiter for a multi-tenant LLM API. Implement the core, then tell me what changes when it's distributed. | Scaling your API with rate limitersHow we built rate limiting capable of scaling to millions of domainsToken-based multi-tenant limits and distributed counting |
| 1 | multi-language code reviewAnthropic 1Hiring-manager call (technical variant): review code examples in different programming languages, detect issues, and identify what the code … | Google Engineering Practices: Code ReviewLanguage-agnostic code review checklist |
| 1 | file system traversalOpenAI 1Practical coding phone screen (CoderPad): algorithms and data-structures problem framed as real work, e.g. traversing a file system; multi-p… | LeetCode 588 Design In-Memory File System / 71 Simplify PathAnthropic Interview Notes (unofficial reconstructed problems)File-system trie; os.walk-based traversal in dedup notes |
| 1 | time-based data structureOpenAI 1Coding problems built on time-based data structures. | LeetCode 981 Time Based Key-Value Store / 1146 Snapshot ArrayTime-indexed values with bisect |
| 1 | versioned data storeOpenAI 1Coding problems built on versioned data stores. | LeetCode 981 Time Based Key-Value Store / 1146 Snapshot ArrayOpenAI Interview Notes (unofficial reconstructed problems)Versioned store; time-based-kv-store |
| 1 | coroutines and concurrencyOpenAI 1Coding that uses coroutines in your chosen language (multithreading, concurrency). | import asyncio: Learn Python's AsyncIO (video series)Python docs: asyncio — coroutines and tasks, synchronization primitivesFluent Python, 2nd EditionCoroutines from event loop up |
| 1 | OOP iterator classesOpenAI 1Coding that tests object-oriented programming concepts: abstract classes, iterator classes, inheritance. | Fluent Python, 2nd EditionLeetCode 341 Flatten Nested List Iterator / 284 Peeking IteratorIterator protocol and ABCs; iterator classes |
| 1 | agentic coding existing codebaseOpenAI 1Agentic coding round (beta): given a prompt and an existing codebase, make changes and add features to a problem scoped too large to solve b… | Best practices for Claude CodeAgentic changes in an existing codebase |
| 1 | Meta AI-assisted coding roundMeta MSL 1AI-assisted coding round (60 min, CoderPad with built-in assistant, multi-file codebase of a few hundred to a few thousand lines): one thema… | Meta's AI-Enabled Coding Interview: How to PrepareMeta AI-enabled round format and rubric |
| 1 | take-home with evalsUnspecified frontier 1Take-home of four to eight hours with a written report, sometimes with AI explicitly allowed; must run with one command, ship evals, and the… | Your AI Product Needs EvalsEvals section for take-home reports |
| 1 | image instruction-file pipelineAnthropic 1Code an image-processing pipeline that reads per-image instruction files and applies ordered transformations (e.g., rotate, invert, crop, co… | Anthropic Interview Notes (unofficial reconstructed problems)Pillow (PIL fork) documentationInstruction-file image pipeline |
| 1 | LRU cache with extensionsAnthropic 1Implement an LRU cache with serialization and evolving constraints. | Anthropic Interview Notes (unofficial reconstructed problems)LeetCode 146 LRU Cache / 460 LFU CacheLRU with serialization and evolving constraints |
| 1 | streaming database progressive specAnthropic 1Implement a streaming database with the given constraints and requirements. | Open Gauntlet: 4-level progressive coding practiceAnthropic Interview Notes (unofficial reconstructed problems)Progressive streaming-DB style levels |
| 1 | in-memory database CodeSignal levelsAnthropic 1Implement an in-memory database in CodeSignal, with each round adding more functionality. | Anthropic Interview Notes (unofficial reconstructed problems)oa-in-memory-database CodeSignal levels |
| 1 | chat interface with classesOpenAI 1Build a chatbot-style chat interface using classes. | OpenAI Interview Notes (unofficial reconstructed problems)chat-bot-refactoring class design |
| 1 | GPU credit management systemOpenAI 1Implement a GPU credit management system. | OpenAI Interview Notes (unofficial reconstructed problems)gpu-credits management |
| 1 | task scheduler in PythonOpenAI 1Design a task scheduler in Python. | LeetCode 1834 Single-Threaded CPU / 621 Task SchedulerPython docs: collections, heapq, bisect, itertools, dataclassesHeap-based Python task scheduler |
| 1 | stateful iterator over iteratorsxAI 1Implement an iterator class that supports next, hasNext, getState, and setState, then extend it to treat a list of iterators as one continuo… | OpenAI Interview Notes (unofficial reconstructed problems)resumable-iterator getState/setState |
| 1 | implement missing fn in LLM codebasexAI 1Given a large, partially implemented LLM/agentic codebase, identify the intended behavior and implement a missing function by reading and re… | vLLM Architecture OverviewnanoGPT (code-reading practice)Infer intended behaviour from LLM codebase structure |
| 1 | beat string to music notationOpenAI 1Convert a string marking whether an instrument is played on each beat into music notation; extend to multiple instruments, then rests. | Exercism Python trackString-to-notation conversion drills |
| 1 | KV store serializer state restoreOpenAI 1Implement a key-value store serializer and deserializer that stores and restores system state, with specific output when state is queried wh… | OpenAI Interview Notes (unofficial reconstructed problems)Python docs: pickle, shelve and json (data persistence)durable-kv-store state restore; serialization choices |
| 1 | human/AI task assignment systemOpenAI 1Write code for a system that assigns tasks to humans and AI, satisfying a provided list of requirements. | OpenAI Interview Notes (unofficial reconstructed problems)labeling-task-scheduler human/AI assignment |
| 1 | iterator 2D and asyncOpenAI 1Build an iterator that outputs values in sequence, then extend it to 2D, then to an async iterator. | LeetCode 341 Flatten Nested List Iterator / 284 Peeking IteratorFluent Python, 2nd EditionPython docs: asyncio — coroutines and tasks, synchronization primitivesNested/2D iterators; async iterators |
| 1 | KV state cloud sync bytesOpenAI 1Efficiently sync key/value state to and from the cloud, using helper methods that convert strings to bytes and back. | Python docs: struct — interpret bytes as packed binary dataProtocol Buffers: EncodingBytes encoding for cloud sync |
| 1 | chess game OOPOpenAI 1Design a chess game. | awesome-low-level-designChess OOD |
| 1 | banking app progressive specAnthropic 1Build the core business logic for a banking application with multiple transaction types. | Anthropic Interview Notes (unofficial reconstructed problems)oa-bank-system transaction types |
| 1 | extend feature unfamiliar codebaseAnthropic 1Extend a practical feature inside an existing codebase with limited documentation. | Working Effectively with Legacy CodeExtend features with seams and tests |
| 1 | stdlib-only data extractionAnthropic 1Python data extraction task restricted to the standard library (no third-party NLP packages), plus ethical trade-off questions about data fo… | Regular Expression HOWTO (Python docs) + csv moduleStdlib regex/csv extraction |
| 1 | messy dataset cleaning pandasAnthropic 1You're given a roughly 100-column sales dataset with missing values, inconsistent formats, and duplicate records. Transform it for downstrea… | Python for Data Analysis, 3E (open access)10 minutes to pandasCleaning missing values and inconsistent formats |
| 1 | compound-rule rate limiterxAI 1Design and implement a rate limiter for a single URL, then extend it to support compound rules across multiple URLs and payload-aware limits… | Scaling your API with rate limitersLeetCode 359 Logger Rate Limiter / 362 Design Hit CounterComposable rules over token buckets and windows |
| 1 | retry with backoff wrapperOpenAI 1Applied coding: write a retry-with-exponential-backoff wrapper for API calls (or parse tokenizer output / implement streaming batching). | Timeouts, retries, and backoff with jitter (Amazon Builders' Library)Exponential backoff with full jitter wrapper |
| 1 | tree message-passing node countOpenAI 1Cluster Node Count & Topology (Tree Messaging) (3 parts) | OpenAI Interview Notes (unofficial reconstructed problems)cluster-count-topology tree messaging |
| 1 | GPU credit ledgerOpenAI 1GPU Credits (3 parts) | OpenAI Interview Notes (unofficial reconstructed problems)gpu-credits three parts |
| 1 | labeling task schedulerOpenAI 1Data Labeling Task Scheduler (3 parts) | OpenAI Interview Notes (unofficial reconstructed problems)labeling-task-scheduler three parts |
| 1 | monster battle simulationOpenAI 1Monster Battle (3 parts) | OpenAI Interview Notes (unofficial reconstructed problems)monster-battle three parts |
| 1 | toy language type inferenceOpenAI 1Type Inference for a Toy Language (2 parts) | OpenAI Interview Notes (unofficial reconstructed problems)Type inferencetype-inference two parts; unification |
| 1 | checkpointing PR code reviewGoogle DeepMind 1Code Review: Ranking the Defects in a Checkpointing Change: A pull request adds the module checkpointing.py below to a training codebase. | Anthropic Interview Notes (unofficial reconstructed problems)Google Engineering Practices: Code Reviewagentic-pr-review defect ranking; review severity |
| 1 | ML experiment config mergeAnthropic 1ML Experiment Configuration System (4 parts / 55 min) | Anthropic Interview Notes (unofficial reconstructed problems)ml-config-system four parts |
| 1 | weighted data batcher checkpointingAnthropic 1Weighted Data Batcher with Checkpointing (3 parts / 55 min) | Anthropic Interview Notes (unofficial reconstructed problems)weighted-data-batcher with checkpointing |
| 1 | friend recommendation systemMeta MSL 1Friend Recommendation System | OpenAI Interview Notes (unofficial reconstructed problems)social-network friend recommendations |
| 1 | codebase bugfix then extendMeta MSL 1AI-enabled coding phase 1: find and fix a non-algorithmic bug in an existing multi-file codebase (e.g. an iteration cap of 10,000 throwing a… | Best practices for Claude CodeWorking Effectively with Legacy CodeFix then extend inside multi-file codebase |
| 1 | cd command symlinksOpenAI 1Unix CD Command with Symbolic Link Resolution: implement cd navigation with ., .., absolute paths and symlinks, detecting cycles. | OpenAI Interview Notes (unofficial reconstructed problems)LeetCode 588 Design In-Memory File System / 71 Simplify Pathcd-command symlinks; Simplify Path |
| 1 | snapshot arrayOpenAI 1Implement a SnapshotArray (set, snap, get by snap_id). | LeetCode 981 Time Based Key-Value Store / 1146 Snapshot ArrayLC 1146 Snapshot Array |
| 1 | toy language interpreterOpenAI 1Toy language interpreter (75-minute): lexer, recursive-descent parser, evaluator. | Crafting Interpreters500 Lines or Less: A Python Interpreter Written in PythonLexer, recursive-descent parser, evaluator |