Search the siteSearch in

Interview paths
Question bank
Coding practice
Study materials
Infrastructure lab

Community

Weekly sessions
Peer practice
Contribute

News

Jobs

Open roles
Market intelligence
Companies
Roles

Feedback

Product feedback
Developer feedback

Checking account…

Home
Question bank

Frontier AI technical interview question bank.

Practice coding, system design, behavioral, and project questions for frontier AI roles.

Company

Loading the question catalog…

Preparing this page…

Your current navigation and account state stay available while the next page loads.

Frontier Rehearsal · Independent practice · Not affiliated with OpenAI, Anthropic, or xAI.

Home Interview questions Company interview processes Company reports Role reports Market intelligence Open roles Python lab Materials How it works About Feedback Uploads Feature requests Terms Privacy 中文 English

Home
Question bank

Frontier AI technical interview question bank.

Practice coding, system design, behavioral, and project questions for frontier AI roles.

Company

FREE PRACTICE

Public demos, always pinned.

Start with either demo. They remain here when you filter or sort.

Free account: sign in to practice the marked Free questions.

Newest published questions; deterministic, never randomized.

Demo 1Number search with delayed feedback
Demo 2Design real-time device monitoring and control

Question bank filters

Reset filters

Question groups

Filter question bank by interview groupAll 104 Coding 45 System design 30 Behavioral 12 Hiring manager 10 Project deep dive 7

Topics

Filter question bank by topic tagsAll 104 Python 44 Algorithms 15 Data structures 19 Concurrency 12 Distributed systems 3 Storage & state 22 Networking 4 Reliability 19 ML / AI 13 Data / SQL 12 Security 2 Product design 19 Communication 31

Independent and not affiliated with the companies represented. Collected material does not predict future interview questions.

94 questions

Sort question bankSortFree first Newest Oldest AI ratio ↓AI ratio ↑

Question bank results

Every question has a public brief with the task, constraints, and follow-ups. Labels describe interactive practice access.

Free practice

Topic tags
Communication

OpenAIBehavioral

Discussion / designResponse prompt

Reported interview periodNot available

Why this work

Two minutes on why this lab's work fits a project you have done.

Practice focusRecruiter round, two minutes. The question is why this work. Use one project. Sentence one: who it was for. Sentence two: the constraint. Sentence three: what you did with your own hands. Sentence four: the skill that transfers. If you reach for a slogan, stop and return to the project. The interviewer asks what you would want to learn in the first month. Answer with one skill, not a mission statement.

Free practice

Topic tags
Storage & state

OpenAISystem design

Discussion / designResponse prompt

Reported interview periodNot available

Shared scratch pad

Two people edit one note and need to see who wrote the last line.

Practice focusSystem design round. Two people share one scratch pad during a review. Start by listing what must be true: neither person loses a line, and both can point at who wrote the last line. Place the note in one store and give each save a version. Walk the case where both save the same version. Then the network drops the second save: the first person must not think the second line landed. Then add a third reader who only watches. Say what that reader sees, what you store, and what you would not build for a note this small.

Free practice

Topic tags
Communication

OpenAIProject deep dive

Discussion / designResponse prompt

Reported interview periodNot available

After launch

Where a design failed after launch, what you measured, and what you changed.

Practice focusProject round. One system, after launch. Start with the design you intended. Then the failure, including who felt it. Then the measurement, including the unit and the window. Then the change. The interviewer asks what you left alone and why. Answer with one thing you refused to change. Then they ask how you would know the failure had returned. Point at the same measurement. Do not switch projects.

Free practice

Topic tags
Communication

OpenAIBehavioralAI-generated

Discussion / designResponse prompt

Reported interview periodNot available

A decision you own

One behavioral story about a decision made when the facts were thin.

Practice focusBehavioral round. Tell one story about a decision you owned when the facts were thin. When it was. What you knew. What you did not know. The call you made. Who you told. What happened after. Stay on that one story.

Free practice

Topic tags
Communication

OpenAIBehavioralAI-generated

Discussion / designResponse prompt

Reported interview periodNot available

A limit you named

A culture conversation about a limit you set on a piece of work.

Practice focusCulture round. Name a limit you set on a piece of work. What the work was. Who asked you to go past the limit. What you said. What you offered instead. Then say how you would check, a week later, that the limit still held.

Free practice

Topic tags
Storage & state
Product design
Communication

OpenAIBehavioralHalf AI-generated

Discussion / designResponse prompt

Reported interview periodNot available

Cross-functional disagreement

Tell one story about a disagreement with a product, research, policy, or operations partner. State the shared goal, the competing constraints, what you changed in your own position, and the decision the group m…

Practice focusPreparation - Write the user or stakeholder, the constraint, and the decision in one sentence each. - Choose one measurement or observable fact that can support the answer. - Name one alternative and the reason it was not selected. Review checklist 1. Did I state the concrete constraint before describing the solution? 2. Did I separate my own decision and work from the team's work? 3. Did I name evidence or a measurement rather than relying on confidence? 4. Can I answer every follow-up without changing the original facts? Staged follow-ups 1.…

Free practice

Topic tags
Storage & state
Communication

OpenAIProject deep diveHalf AI-generated

Discussion / designResponse prompt

Reported interview periodNot available

Cost and latency project

Walk through a project where you reduced latency or cost without weakening correctness. Cover the baseline, bottleneck, alternatives, the change you owned, rollout safeguards, and measured result.

Free practice

Topic tags
Reliability
Product design

OpenAISystem designHalf AI-generated

Discussion / designResponse prompt

Reported interview periodNot available

Tool execution platform

Design a platform where an AI assistant invokes approved tools on behalf of users. Include tool registration, argument validation, identity and consent, isolated execution, timeouts, idempotency, audit logs, an…

New public brief

Topic tags
Storage & state
Product design

OpenAISystem design

Discussion / designResponse prompt

Reported interview periodSep 2026

Design real-time device monitoring and control

Design a service that ingests telemetry from many devices, detects meaningful state changes, and triggers control actions according to configurable rules.

Practice focusCover the ingestion API, event schema, message transport, stream processing, state storage, rule evaluation, and command-delivery path. Explain how the system handles bursty traffic, duplicate and out-of-order events, intermittent device connectivity, failed control commands, and service recovery. Define the latency and consistency guarantees you would offer, the metrics needed to operate the system, and the safeguards that prevent unstable feedback loops. Follow-ups 1. Support devices that remain offline for days and later upload buffered telemetry. 2.…

New public brief

Topic tags
Python

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodSep 2026

Number search with delayed feedback

A hidden integer lies in the inclusive range 1..n, where n >= 1. The test fixture supplies a check(x) callback. Your code must not implement the callback and is not given the hidden integer.

Practice focusEach call to check(x) submits a new integer guess in 1..n , but returns the comparison result for the previous submitted guess: - the first call returns None ; - -1 means the previous guess was too high; - 0 means the previous guess was correct; and - 1 means the previous guess was too low. Implement: Return the hidden integer. The fixture validates the answer and permits at most 2 n.bit length() calls to check . Track which guess each delayed response describes; do not apply a response to the newly submitted guess.…

New public brief

Topic tags
Python
Algorithms
Data / SQL

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodJan 2026–Sep 2026

Implement 1-nearest-neighbor classification as a linear layer

You are given a matrix of labeled training examples and a batch of query examples. Implement squared-Euclidean 1-nearest-neighbor classification, then show that the same nearest-neighbor choice can be expressed…

Practice focusScenario You are given a matrix of labeled training examples and a batch of query examples. Implement squared-Euclidean 1-nearest-neighbor classification, then show that the same nearest-neighbor choice can be expressed as the output of an affine layer. For a query vector x and training vector x i : -||x - x i||² = 2x·x i - ||x i||² - ||x||² The final term is constant across all training examples for a fixed query, so it does not affect the winning index. This yields an affine score matrix with W = 2 X train.T and b i = -||x i||² .…

New public brief

Topic tags
Python
Algorithms

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodNov 2025–Sep 2026

Infection Spread on a Grid

A grid contains healthy cells (0) and infected cells (1). Each day, infection spreads simultaneously from infected cells to their four orthogonal healthy neighbors. Return the number of days until every suscept…

Practice focusRequirements - All initially infected cells spread at the same time. - Return 0 when no healthy cell exists initially. - Do not allow newly infected cells to spread again during the same day. - Handle empty grids and validate rectangular input. - Target O(rows cols) time and space proportional to the active frontier. In the second stage, cells marked 2 are immune: they never change and block propagation. Follow-ups 1. Add recovery after a configurable number of days. 2. Require at least q infected neighbors before a cell changes. 3.…

New public brief

Topic tags
Python
Data structures

OpenAICodingHalf AI-generated

Python codingWebAssembly + public tests

Reported interview periodJan 2026–Sep 2026

Aggregate Labels from Noisy Annotators

You receive labels from multiple annotators, with missing entries. Infer one binary label per item, identify annotators who systematically disagree with their peers, remove them, and recompute the labels.

Practice focusAssumptions Rows are annotators, columns are items, labels are 0 , 1 , or -1 for missing. Majority ignores missing entries and ties resolve to 1 . An annotator is bad when agreement with provisional labels on overlapping items is strictly below 0.5 . If gold is supplied, also return accuracy rounded to a stated precision. Requirements - Define behavior for items with no labels and annotators with no overlap. - Avoid treating missing values as votes. - Recompute the final majority after filtering. - State complexity and numerical representation.…

Public brief

Topic tags
Algorithms

OpenAICodingAI-generated

Talk-through codingNo local grader

Reported interview periodSep 2026

Deal and rank three-card poker hands

Implement the core of a three-card poker game. Cards are dealt to seated players one card at a time over three passes, then every hand is ranked and the winner or winners are returned.

Practice focusScenario Implement the core of a three-card poker game. Cards are dealt to seated players one card at a time over three passes, then every hand is ranked and the winner or winners are returned. Ranks increase from 2 through A . A deck contains one card for every rank and suit combination, and a card may be dealt at most once. Assumed interface The exact public API is intentionally flexible. One coherent option is: State any alternative interface before implementation.…

Public brief

Topic tags
Python

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodApr 2026–Sep 2026

Expiring compute-credit ledger

Implement a ledger for compute-credit grants and debits whose events may arrive out of order.

Practice focusEach grant has a unique ID, amount, activation timestamp, and expiration duration. A grant is available through its expiration boundary. subtract(amount, timestamp) consumes credits that expire soonest. get balance(timestamp) returns the balance that would have been available at that point in time by evaluating all applicable events in chronological order. Define deterministic behavior for duplicate IDs, insufficient balance, equal timestamps, and late-arriving events. Preserve enough history to answer past-balance queries correctly. Follow-ups 1.…

Public brief

Topic tags
Storage & state
Product design

OpenAISystem designAI-generated

Discussion / designResponse prompt

Reported interview periodFeb 2026–Sep 2026

Design a browser IDE for many users

Design a browser-based development environment where users can open persistent workspaces, edit files, run programs, and interact with a live terminal.

Practice focusCover workspace storage, process isolation, scheduling, standard-output streaming, terminal input, reconnect after refresh, dependency setup, and file synchronization. Explain how you prevent one user’s code from affecting another user or the host system. Address cold-start latency, idle-workspace suspension, resource quotas, long-running jobs, failure recovery, observability, and the trade-off between warm capacity and infrastructure cost. Follow-ups 1. Support collaborative editing without coupling file consistency to terminal I/O. 2.…

Public brief

Topic tags
Concurrency
Reliability
Product design

OpenAISystem designAI-generated

Discussion / designResponse prompt

Reported interview periodApr 2026–Sep 2026

Design a Long-Running Video Generation Platform

Design a service where users submit prompts for video generation. Jobs consume scarce accelerators, may run for minutes, can be cancelled, and must survive worker failure.

Practice focusRequirements - Define APIs for submit, status, progress streaming, cancel, retry, and result retrieval. - Queue and schedule jobs under per-user quotas, priorities, and accelerator constraints. - Use leases or heartbeats so abandoned work becomes recoverable. - Make retries safe, preserve checkpoints, and avoid publishing partial or duplicate results. - Store inputs and outputs securely with retention and content-safety controls. - Provide honest progress despite phases with uncertain duration.…

Public brief

Topic tags
Communication

OpenAIProject deep diveAI-generated

Discussion / designResponse prompt

Reported interview periodDec 2025–Sep 2026

Technical project deep dive

Lead a structured discussion of one technically substantial system you personally helped build. Choose a project with real constraints, meaningful decisions, and at least one outcome you can quantify. The goal …

Practice focusScenario Lead a structured discussion of one technically substantial system you personally helped build. Choose a project with real constraints, meaningful decisions, and at least one outcome you can quantify. The goal is to demonstrate engineering judgment and ownership, not to present a polished product pitch. Suggested structure 1. Context: Who used the system, what problem mattered, and what was explicitly out of scope? 2. Constraints: Scale, latency, reliability, privacy, staffing, deadlines, legacy dependencies, or cost. 3.…

Public brief

Topic tags
Python
Data structures
Concurrency
Networking

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodFeb 2026–Sep 2026

IPv4 and CIDR Iterator

Implement an iterator that walks IPv4 addresses from a seed, either forward, backward, or within the seed's CIDR block.

Practice focusRequirements - Accept a plain IPv4 address or CIDR such as 10.2.3.4/24 . - Include the seed as the first result. - Forward iteration ends at 255.255.255.255 or the block broadcast address. - Backward iteration ends at 0.0.0.0 or the block network address. - Validate octets, prefix lengths, and direction; raise StopIteration after exhaustion. - Use integer address arithmetic rather than incrementing dotted components as strings. State whether network and broadcast addresses are included; for this problem, include both boundaries. Follow-ups 1.…

Public brief

Topic tags
Concurrency
Storage & state
Security

OpenAISystem designAI-generated

Discussion / designResponse prompt

Reported interview periodFeb 2026–Sep 2026

Design an Online Chess Service

Design a service that matches two players, maintains authoritative legal game state, supports clocks, and allows games to be reviewed later.

Practice focusRequirements - Define APIs or real-time events for matchmaking, moves, resignations, draw offers, reconnects, and spectators. - The server—not the client—validates turns, legal moves, check, checkmate, stalemate, and special rules. - Make duplicate or reordered client messages idempotent. - Maintain fair clocks despite network delay and disconnects. - Persist an append-only move history and enough state for fast reconnect and review. - State scale assumptions, latency goals, and consistency boundaries.…

Public brief

Topic tags
Python
Data structures
Concurrency
Storage & state
Reliability

OpenAICodingHalf AI-generated

Python codingWebAssembly + public tests

Reported interview periodAug 2026–Sep 2026

Leased Work Queue with Dead-Letter Handling

Implement an in-memory queue whose jobs are reserved with expiring leases. A failed worker must not hold a job forever, and jobs that exceed their retry budget move to a dead-letter state.

Practice focusRequirements and assumptions - Jobs move among ready , reserved , completed , and dead . - reserve increments attempts and returns {item id, token} plus an internal deadline. - On explicit failure or lease deadline <= now , return the job to ready unless attempts reached max attempts ; then mark it dead. - Completion or failure with a stale token returns False without changing state. - Queue ready jobs deterministically and reject duplicate active IDs. - Validate positive limits and use the explicit clock supplied by callers. Follow-ups 1.…

Public brief

Topic tags
Product design

OpenAISystem design

Discussion / designResponse prompt

Reported interview periodJan 2026–Sep 2026

Design hold-then-charge payments

Design a payment platform with two operations:

Practice focus- hold synchronously requests authorization from a downstream card network and must return approval or rejection within a 500 ms p95 target. - charge references an approved hold, durably records the request, acknowledges it, and includes it in a later settlement batch. Assume 10 million holds and 8 million charges per day, peaks near 300 requests per second, 5–10 downstream networks, a two-hour nightly settlement window, and a seven-day hold lifetime.…

Public brief

Topic tags
Python
ML / AI

OpenAICodingHalf AI-generated

Python codingWebAssembly + public tests

Reported interview periodNov 2025–Sep 2026

Debug a minimal transformer computation

A small NumPy model produces incorrect attention outputs and fails to learn. Implement corrected versions of four operations used by a decoder-style model. The model dimension is small and attention uses one he…

Practice focusScenario A small NumPy model produces incorrect attention outputs and fails to learn. Implement corrected versions of four operations used by a decoder-style model. The model dimension is small and attention uses one head, so focus on tensor semantics rather than framework APIs. Interface Requirements pos embed Return token representations with positional embeddings added elementwise. Shapes must match or broadcast unambiguously. attention - Compute scaled dot-product attention using a scale of 1 / sqrt(d k) .…

Public brief

Topic tags
Python
Data structures
Distributed systems

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodOct 2025–Sep 2026

Aggregate Count and Topology over a Machine Tree

Machines form a rooted tree and may communicate only with their parent or children. Implement message handling so the root can count all machines and reconstruct a deterministic topology string.

Practice focusRequirements - A count request propagates to children; a leaf replies 1 ; an internal node replies with one plus all child counts. - A topology request produces id(child1,child2,...) ; a leaf is id[] . - Only the root publishes the final result. - Correlate replies with operation type and wait for every expected child exactly once. - Do not assume child replies arrive in order; sort by child ID for deterministic topology. - Avoid blocking the message handler and state behavior for duplicate requests.…

Previous2 of 4Next

CONTENT BOUNDARY

Independent frontier AI interview practice. Includes adaptations of collected interview material and original exercises; not official company questions and not affiliated with the companies represented. Collected material does not predict future interview questions. Report a correction.

Loading the question catalog…

FREE PRACTICE

Public demos, always pinned.

Start with either demo. They remain here when you filter or sort.

Free account: sign in to practice the marked Free questions.

Newest published questions; deterministic, never randomized.

Demo 1Number search with delayed feedback
Demo 2Design real-time device monitoring and control

Question bank filters

Reset filters

Question groups

Filter question bank by interview groupAll 104 Coding 45 System design 30 Behavioral 12 Hiring manager 10 Project deep dive 7

Topics

Independent and not affiliated with the companies represented. Collected material does not predict future interview questions.

94 questions

Sort question bankSortFree first Newest Oldest AI ratio ↓AI ratio ↑

Question bank results

Every question has a public brief with the task, constraints, and follow-ups. Labels describe interactive practice access.

Free practice

Topic tags
Communication

OpenAIBehavioral

Discussion / designResponse prompt

Reported interview periodNot available

Why this work

Two minutes on why this lab's work fits a project you have done.

Free practice

Topic tags
Storage & state

OpenAISystem design

Discussion / designResponse prompt

Reported interview periodNot available

Shared scratch pad

Two people edit one note and need to see who wrote the last line.

Free practice

Topic tags
Communication

OpenAIProject deep dive

Discussion / designResponse prompt

Reported interview periodNot available

After launch

Where a design failed after launch, what you measured, and what you changed.

Free practice

Topic tags
Communication

OpenAIBehavioralAI-generated

Discussion / designResponse prompt

Reported interview periodNot available

A decision you own

One behavioral story about a decision made when the facts were thin.

Free practice

Topic tags
Communication

OpenAIBehavioralAI-generated

Discussion / designResponse prompt

Reported interview periodNot available

A limit you named

A culture conversation about a limit you set on a piece of work.

Free practice

Topic tags
Storage & state
Product design
Communication

OpenAIBehavioralHalf AI-generated

Discussion / designResponse prompt

Reported interview periodNot available

Cross-functional disagreement

Free practice

Topic tags
Storage & state
Communication

OpenAIProject deep diveHalf AI-generated

Discussion / designResponse prompt

Reported interview periodNot available

Cost and latency project

Walk through a project where you reduced latency or cost without weakening correctness. Cover the baseline, bottleneck, alternatives, the change you owned, rollout safeguards, and measured result.

Free practice

Topic tags
Reliability
Product design

OpenAISystem designHalf AI-generated

Discussion / designResponse prompt

Reported interview periodNot available

Tool execution platform

New public brief

Topic tags
Storage & state
Product design

OpenAISystem design

Discussion / designResponse prompt

Reported interview periodSep 2026

Design real-time device monitoring and control

Design a service that ingests telemetry from many devices, detects meaningful state changes, and triggers control actions according to configurable rules.

New public brief

Topic tags
Python

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodSep 2026

Number search with delayed feedback

A hidden integer lies in the inclusive range 1..n, where n >= 1. The test fixture supplies a check(x) callback. Your code must not implement the callback and is not given the hidden integer.

New public brief

Topic tags
Python
Algorithms
Data / SQL

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodJan 2026–Sep 2026

Implement 1-nearest-neighbor classification as a linear layer

New public brief

Topic tags
Python
Algorithms

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodNov 2025–Sep 2026

Infection Spread on a Grid

New public brief

Topic tags
Python
Data structures

OpenAICodingHalf AI-generated

Python codingWebAssembly + public tests

Reported interview periodJan 2026–Sep 2026

Aggregate Labels from Noisy Annotators

Public brief

Topic tags
Algorithms

OpenAICodingAI-generated

Talk-through codingNo local grader

Reported interview periodSep 2026

Deal and rank three-card poker hands

Implement the core of a three-card poker game. Cards are dealt to seated players one card at a time over three passes, then every hand is ranked and the winner or winners are returned.

Public brief

Topic tags
Python

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodApr 2026–Sep 2026

Expiring compute-credit ledger

Implement a ledger for compute-credit grants and debits whose events may arrive out of order.

Public brief

Topic tags
Storage & state
Product design

OpenAISystem designAI-generated

Discussion / designResponse prompt

Reported interview periodFeb 2026–Sep 2026

Design a browser IDE for many users

Design a browser-based development environment where users can open persistent workspaces, edit files, run programs, and interact with a live terminal.

Public brief

Topic tags
Concurrency
Reliability
Product design

OpenAISystem designAI-generated

Discussion / designResponse prompt

Reported interview periodApr 2026–Sep 2026

Design a Long-Running Video Generation Platform

Design a service where users submit prompts for video generation. Jobs consume scarce accelerators, may run for minutes, can be cancelled, and must survive worker failure.

Public brief

Topic tags
Communication

OpenAIProject deep diveAI-generated

Discussion / designResponse prompt

Reported interview periodDec 2025–Sep 2026

Technical project deep dive

Public brief

Topic tags
Python
Data structures
Concurrency
Networking

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodFeb 2026–Sep 2026

IPv4 and CIDR Iterator

Implement an iterator that walks IPv4 addresses from a seed, either forward, backward, or within the seed's CIDR block.

Public brief

Topic tags
Concurrency
Storage & state
Security

OpenAISystem designAI-generated

Discussion / designResponse prompt

Reported interview periodFeb 2026–Sep 2026

Design an Online Chess Service

Design a service that matches two players, maintains authoritative legal game state, supports clocks, and allows games to be reviewed later.

Public brief

Topic tags
Python
Data structures
Concurrency
Storage & state
Reliability

OpenAICodingHalf AI-generated

Python codingWebAssembly + public tests

Reported interview periodAug 2026–Sep 2026

Leased Work Queue with Dead-Letter Handling

Implement an in-memory queue whose jobs are reserved with expiring leases. A failed worker must not hold a job forever, and jobs that exceed their retry budget move to a dead-letter state.

Public brief

Topic tags
Product design

OpenAISystem design

Discussion / designResponse prompt

Reported interview periodJan 2026–Sep 2026

Design hold-then-charge payments

Design a payment platform with two operations:

Public brief

Topic tags
Python
ML / AI

OpenAICodingHalf AI-generated

Python codingWebAssembly + public tests

Reported interview periodNov 2025–Sep 2026

Debug a minimal transformer computation

Public brief

Topic tags
Python
Data structures
Distributed systems

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodOct 2025–Sep 2026

Aggregate Count and Topology over a Machine Tree

Machines form a rooted tree and may communicate only with their parent or children. Implement message handling so the root can count all machines and reconstruct a deterministic topology string.

Previous2 of 4Next

CONTENT BOUNDARY