Choose a company and role, or a future roadmap, to start your interview.
Loading interviews…
Company interviews
Historical & published processes
Select a company and role to start. Each interview includes behavioral, technical and system design questions, with adaptive follow-ups. Published process guides are below; ↔ means the count varies or is unpublished. Talk naturally, pause whenever you need, and review all scores and feedback after you finish.
Big tech
Engineering at a global scale.
GPublished process
Google
Search · Cloud · Platforms
Technical candidate guide
Coding & design interviews
↔TechnicalCoding & problem solving
↔SystemsDesign discussions · role dependent
Process context +
The virtual interview guide covers coding and design formats, but does not publish a fixed round breakdown. Ask your recruiter for your specific loop.
A Monzo engineer's historical account: recruiter call, take-home exercise, code-review discussion, then systems and behavioural interviews. The technical count includes the take-home assessment. This is one account, not a universal current loop.
Published generalist process: recruiter screen, take-home, two technical interviews, hiring manager, then debrief and team matching. The handbook describes an India-only pilot; specialist roles use a different process.
A historical breakdown is on our research list. Start an interview for this company below.
Process context +
A sourced historical round breakdown has not been added yet. You can still choose company-context practice in the interview setup above.
No sourced round counts yet.
AI labs
Research meets production engineering.
OPublished process
OpenAI
AI research · Products
Published engineering hiring guide
4–6 final interviews · format varies
↔TechnicalEngineering solutions, code & tests
↔CollaborationCommunication & problem solving
Process context +
The published guide describes 4–6 hours with 4–6 people in the final stage. It does not assign fixed counts to each interview type. The guide also describes introductory calls and skills assessments.
The careers page describes live coding for technical roles and discussions of experience and motivation. A fixed round count is not published; formats depend on the role.
Snapshots describe the source’s role and scope, including assessments. Processes can change by team, seniority and location.
Future interview roadmaps
✧ Research outlook
Emerging technology, translated into a practical learning path. These are editorial hypotheses grounded in foundational research, not historical interview reports or validated predictions.
01 / RESEARCH ROADMAP↗
Agentic AI
Agents that can act safely
Tool usePermissionsEvaluation
As models take actions through tools, engineering interviews could explore how you bound autonomy, recover from failures and measure task success.
Explore the roadmap +
Understand the loop
Trace how reasoning, tool calls and observations change an agent's next action.
Design the boundaries
Plan least-privilege tools, approval points, idempotency and recovery from partial failures.
Evaluate end to end
Create a task suite that measures successful outcomes, unsafe actions, cost and latency.
POSSIBLE FUTURE QUESTION · ORIGINAL EXAMPLE
“An AI support agent can issue refunds. How would you constrain its permissions and recover safely when a tool times out?”
AI products need answers grounded in changing information. A future design discussion could test retrieval quality, access control and evaluation beyond a convincing demo.
Explore the roadmap +
Build the foundation
Understand embeddings, chunking, indexing and the difference between retrieval and generation failures.
Make it dependable
Design document permissions, freshness checks, citations and abstention when evidence is missing.
Prove it works
Separate retrieval metrics from answer faithfulness; test stale, missing and contradictory evidence.
POSSIBLE FUTURE QUESTION · ORIGINAL EXAMPLE
“Design an internal knowledge assistant. How would you keep answers current and prevent it from revealing documents a user cannot access?”
Serving models introduces memory, throughput and latency trade-offs. Systems interviews could ask you to reason about these constraints alongside traditional scaling concerns.
Explore the roadmap +
Learn the bottlenecks
Explain prefill, decoding and why the KV cache consumes memory as context and concurrency grow.
Compare the trade-offs
Explore paged memory, batching and admission control under a fixed GPU budget.
Design for real traffic
Propose a load test measuring time to first token, tail latency, throughput and cost per request.
POSSIBLE FUTURE QUESTION · ORIGINAL EXAMPLE
“Your model-serving service runs out of GPU memory during traffic spikes. How would you improve concurrency while meeting a tail-latency target?”
Research informs the preparation topics. It does not establish that any company will ask these questions. Paper dates refer to the original research, not a newly measured hiring trend.
Independent preparation. No company endorsement. Practice questions are generated, not actual company questions.Your account