Frontier AI Technical Interview Question Bank

Topic tags
Python
Data / SQL

Reported interview periodNov 2025

Build an in-memory spreadsheet with formulas

Implement a small spreadsheet that stores numeric cells and formulas. A formula may reference other cells, and reads must reflect the latest values after any update.

Practice focusScenario Implement a small spreadsheet that stores numeric cells and formulas. A formula may reference other cells, and reads must reflect the latest values after any update. Support formulas consisting of cell references, numeric literals, parentheses, and the arithmetic operators + , - , , and / . Cell names contain uppercase letters followed by a positive row number, such as A1 or BC12 . Interface Requirements - setCell(key, value) stores either a numeric string such as "12.5" or a formula beginning with = .…

Free practice

Topic tags
Python
Storage & state
Data / SQL

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodNov 2025

In-Memory Relational Tables and Queries

Implement a small in-memory database supporting table creation, row insertion, filtering, projection, and ordering.

Practice focusRequirements - Reject duplicate tables and rows containing unknown columns; define defaults for missing columns. - where , when supplied, is a predicate over a row. - Projection returns only requested existing columns and must not expose mutable internal rows. - order by=(columns, ascending) sorts lexicographically by listed columns with deterministic ties. - Define null and mixed-type comparison behavior. - Preserve insertion order when no ordering is requested. Follow-ups 1. Add joins and grouped aggregations. 2.…

Free practice

Topic tags
Data / SQL

OpenAISystem designAI-generated

Discussion / designResponse prompt

Reported interview periodOct 2025

Design a nearby-places search service

Design a consumer service that returns nearby businesses with ratings, opening hours, and recent reviews. A user supplies a location, an optional query or category, and filters such as “open now.” Results shoul…

Practice focusScenario Design a consumer service that returns nearby businesses with ratings, opening hours, and recent reviews. A user supplies a location, an optional query or category, and filters such as “open now.” Results should feel fresh and should rank useful places ahead of merely close places. Assume a metropolitan launch with tens of millions of businesses globally, high read traffic, and continuous updates from business owners and reviewers. Core requirements - Search by latitude/longitude and radius, optionally constrained by text or category.…

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 .…

Public brief

Topic tags
Python
Data / SQL

OpenAICoding

Python codingWebAssembly + public tests

Reported interview periodApr 2026–Aug 2026

Online softmax entropy

Implement numerically stable softmax, log-softmax, and entropy for a vector of logits. The standard entropy is -Σ p log p.

Practice focusThen implement a block-wise variant that processes the logits in bounded-size chunks without losing numerical stability. The block-wise result must match the full-vector computation within floating-point tolerance. Handle very large and very small logits, empty inputs according to the provided contract, and both supported entropy-sign conventions. Explain why subtracting the maximum logit prevents overflow and how partial statistics combine across blocks. Follow-ups 1. Return gradients with respect to the logits. 2.…

Public brief

Topic tags
Python
Data structures
Reliability
Data / SQL

Coding exerciseCodingAI-generated

Runnable coding50 min

Reported interview periodN/A

Dependency-ready scheduler

Turn a dependency graph into stable ready sets, execution layers, and failure impact.

Practice focusPython, Graphs, Stable ordering, Cycle detection, Failure analysis through a runnable coding exercise.

Public brief

Topic tags
Python
Data / SQL

Coding exerciseCodingAI-generated

Runnable coding50 min

Reported interview periodN/A

Evaluation slice aggregator

Aggregate pass rates, enforce minimum support, and compare evaluation runs without hiding weak slices.

Practice focusPython, Aggregation, Evaluation, Statistical hygiene, Determinism through a runnable coding exercise.

Public brief

Topic tags
Data / SQL
Communication

Behavioral practiceBehavioralAI-generated

Written response35–50 min

Reported interview periodN/A

Build one adaptable decision story

Turn one real experience into a concise story, then answer increasingly specific follow-ups without inventing evidence.

Practice focusEvidence selection, Concise communication, Conflict, Reflection through a written response exercise.

Public brief

Topic tags
Reliability
ML / AI
Data / SQL

System designSystem designAI-generated

Written response70–90 min

Reported interview periodN/A

Design an evaluation artifact registry

Design a reproducible registry connecting datasets, evaluator configurations, model outputs, and reviewed results.

Practice focusData lineage, Versioning, Idempotency, Access control, Reproducibility through a written response exercise.

Public brief

Topic tags
Algorithms
Data structures
Storage & state
Reliability
Data / SQL

OpenAISystem designAI-generated

Written response45–60 min

Reported interview periodN/A

A mutable indexed sequence that must restart exactly

Design a sequence that stays fast to index, accepts mutation, and can crash at any point without losing the state a reader already saw. Then make the restart path provable rather than plausible.

Practice focusData structure design, Complexity analysis, Durability, Crash recovery through a written response exercise.

Public brief

Topic tags
Storage & state
Data / SQL

System designSystem design

Written response45 minutes

Reported interview periodN/A

Weekly score board

Store a score per model run and show which run is ahead this week.

Practice focusRead path, Write path through a written response exercise.

Frontier AI technical interview question bank.

Frontier AI technical interview question bank.

Build an in-memory spreadsheet with formulas

In-Memory Relational Tables and Queries

Design a nearby-places search service

Implement 1-nearest-neighbor classification as a linear layer

Online softmax entropy

Dependency-ready scheduler

Evaluation slice aggregator

Build one adaptable decision story

Design an evaluation artifact registry

A mutable indexed sequence that must restart exactly

Weekly score board

Build an in-memory spreadsheet with formulas

In-Memory Relational Tables and Queries

Design a nearby-places search service

Implement 1-nearest-neighbor classification as a linear layer

Online softmax entropy

Dependency-ready scheduler

Evaluation slice aggregator

Build one adaptable decision story

Design an evaluation artifact registry

A mutable indexed sequence that must restart exactly

Weekly score board