Implement an LRU Cache (LeetCode 146) and explain your design decisions; expect follow-ups on thread-safety / TTL expiry.

Combine a hash map (O(1) lookup) with a doubly linked list (O(1) eviction); discuss the node struct, then walk get/put. For follow-ups, add a lock or a lazy-expiry timestamp rather than rewriting. State complexity up front. Knowing the canonical hashmap+DLL pattern cold, clean pointer manipulation without an IDE, and handling edge cases (capacity 1, updating an existing key). It is reported as 'extremely common' at Microsoft.

Find the length of the longest substring without repeating characters (LeetCode 3).

Sliding window with a last-seen index map; advance the right pointer, jump the left pointer past the previous occurrence. O(n) time, O(min(n, charset)) space. Verbalize why brute force O(n^2) is wasteful. Recognizing the sliding-window pattern, correct window-shrink logic, and ASCII vs Unicode charset assumptions. One of the most consistently reported Microsoft problems.

Search in a rotated sorted array — binary search a rotated array for a target (LeetCode 33).

Modified binary search: at each midpoint decide which half is sorted, then test whether the target lies within that sorted half to pick the side. Be precise with <=/< boundary conditions; handle duplicates if asked. Comfort adapting binary search invariants, off-by-one discipline, and reasoning about which half is monotonic. Reported in a 2025 Microsoft write-up as 'a very standard, famous question.'

Design a cloud file-sync service like OneDrive/Dropbox: upload, sync across multiple devices in near real-time, version history, and conflict resolution.

Start with a 3-tier core (client, metadata service, blob storage), then layer chunked uploads, a sync/notification service, and a metadata DB keyed by file version. Drive the conversation to conflict resolution (last-writer-wins vs vector clocks) and dedup via content-hash chunking. Cloud-native instincts, handling the hard part (conflict resolution + incremental sync), and not over-engineering before the core works. Reported repeatedly as a Microsoft favorite; phrase it in Azure vocabulary.

Design the chat service for Microsoft Teams: real-time messaging and presence for millions of concurrent users.

Persistent connections (WebSocket) via a gateway/fan-out layer, a message store partitioned by conversation, a pub/sub backbone for delivery, and a separate presence service with heartbeats + TTL. Discuss ordering, delivery/read receipts, and group-chat fan-out. Reasoning about connection management at scale, ordering guarantees, and the presence-service design specifically. Reported as a common Microsoft HLD; depth on real-time delivery separates a level-up from a downlevel.

Design a URL shortener — with a deep dive on multi-region deployment and the ID-generation process.

Base62-encode a unique 64-bit ID; the interesting part is ID generation: discuss a Snowflake-style scheme (timestamp + machine/region bits + sequence) to avoid cross-region collisions, plus caching reads and a KV store for the mapping. Address consistency vs latency across regions. Going beyond the toy version into distributed ID generation (Snowflake, Base62) and multi-region consistency. Reported verbatim in a 2025 L61/62 loop as the HLD round.

Tell me about a time you had a conflict with someone. How did you resolve it, and what did you learn?

Use STAR with a real peer/cross-team conflict; focus on understanding the other side, the concrete action you took to align, and a learning that shows growth — not on being right. Keep it specific and own your part. Collaboration, empathy, and a growth-mindset framing (challenge as a learning event). The 'what did you learn' tail is doing the scoring; canned 'we just talked it out' answers fail.

Describe a time your project failed (or you made a mistake). What would you have done differently?

Pick a genuine failure you owned, explain the decision and its impact honestly, then the systemic lesson and how you applied it later. Avoid blaming others or disguising a humblebrag as a failure. Accountability, self-awareness, and learning over defensiveness. A defensive answer here is exactly what the AA round will hammer on later, so candor matters.

Tell me about a time you influenced or convinced someone without having authority over them.

STAR: show how you used data, prototypes, or relationships (not title) to change a decision; quantify the outcome. Emphasize listening and tailoring the argument to their incentives. Influence, communication, and cross-team leadership signal — core to senior Microsoft levels. They want evidence you move others through reasoning, not escalation.

How do you evaluate an ML model, and when would you look beyond accuracy?

Tie metrics to the problem: precision/recall/F1 and ROC-AUC for imbalanced classification, calibration where probabilities matter, plus a held-out/cross-validation strategy and guarding against leakage. Connect metric choice to the business cost of each error type. Practical evaluation judgment (not just 'accuracy'), awareness of class imbalance and validation hygiene. Maps to Microsoft DS classification-metrics questions.

Microsoft Interview Questions

Q: Design a cloud file-sync service like OneDrive/Dropbox: upload, sync across multiple devices in near real-time, version history, and conflict resolution.

Start with a 3-tier core (client, metadata service, blob storage), then layer chunked uploads, a sync/notification service, and a metadata DB keyed by file version. Drive the conversation to conflict resolution (last-writer-wins vs vector clocks) and dedup via content-hash chunking. Cloud-native instincts, handling the hard part (conflict resolution + incremental sync), and not over-engineering before the core works. Reported repeatedly as a Microsoft favorite; phrase it in Azure vocabulary.

Q: Design the chat service for Microsoft Teams: real-time messaging and presence for millions of concurrent users.

Persistent connections (WebSocket) via a gateway/fan-out layer, a message store partitioned by conversation, a pub/sub backbone for delivery, and a separate presence service with heartbeats + TTL. Discuss ordering, delivery/read receipts, and group-chat fan-out. Reasoning about connection management at scale, ordering guarantees, and the presence-service design specifically. Reported as a common Microsoft HLD; depth on real-time delivery separates a level-up from a downlevel.

Q: Design a URL shortener — with a deep dive on multi-region deployment and the ID-generation process.

Base62-encode a unique 64-bit ID; the interesting part is ID generation: discuss a Snowflake-style scheme (timestamp + machine/region bits + sequence) to avoid cross-region collisions, plus caching reads and a KV store for the mapping. Address consistency vs latency across regions. Going beyond the toy version into distributed ID generation (Snowflake, Base62) and multi-region consistency. Reported verbatim in a 2025 L61/62 loop as the HLD round.

Q: Tell me about a time you had a conflict with someone. How did you resolve it, and what did you learn?

Use STAR with a real peer/cross-team conflict; focus on understanding the other side, the concrete action you took to align, and a learning that shows growth — not on being right. Keep it specific and own your part. Collaboration, empathy, and a growth-mindset framing (challenge as a learning event). The 'what did you learn' tail is doing the scoring; canned 'we just talked it out' answers fail.

Q: Describe a time your project failed (or you made a mistake). What would you have done differently?

Pick a genuine failure you owned, explain the decision and its impact honestly, then the systemic lesson and how you applied it later. Avoid blaming others or disguising a humblebrag as a failure. Accountability, self-awareness, and learning over defensiveness. A defensive answer here is exactly what the AA round will hammer on later, so candor matters.

Q: Tell me about a time you influenced or convinced someone without having authority over them.

STAR: show how you used data, prototypes, or relationships (not title) to change a decision; quantify the outcome. Emphasize listening and tailoring the argument to their incentives. Influence, communication, and cross-team leadership signal — core to senior Microsoft levels. They want evidence you move others through reasoning, not escalation.

Q: How do you evaluate an ML model, and when would you look beyond accuracy?

Tie metrics to the problem: precision/recall/F1 and ROC-AUC for imbalanced classification, calibration where probabilities matter, plus a held-out/cross-validation strategy and guarding against leakage. Connect metric choice to the business cost of each error type. Practical evaluation judgment (not just 'accuracy'), awareness of class imbalance and validation hygiene. Maps to Microsoft DS classification-metrics questions.

Microsoft's SWE loop is a recruiter/HM screen + (often) a Codility/HackerRank online assessment, then a 4-5 interview "loop" of 2-4 coding rounds, 1-2 design rounds (LLD and/or HLD, weighted heavier from L61+), and behavioral threaded throughout; strong loops end in the distinctive "As Appropriate" (AA) round with a senior leader who holds veto power. Coding is done in CoderPad/Teams with no syntax highlighting, so clean code and thinking aloud matter.

24 real questions across 4 rounds

Reported via LeetCode Microsoft company tag + 2025 candidate write-ups on LeetCode Discuss (SDE-2 'Recent questions 2025 / Consolidated' thread, L63 'Selected' experience, L61/62 experience), corroborated by HelloInterview L63 guide and IGotAnOffer. Note: Microsoft uses CoderPad/Teams without syntax highlighting.

7 more coding questions Microsoft actually asks, each with how to approach it and exactly what the interviewer is evaluating. Unlock the full set across every round.

Continue with Google

Or run a Microsoft mock with the AI interviewer. It asks questions like these, follows up on your answers, and tells you exactly what to fix. The only score that means anything is the one on your real answers.

Continue with Google

Keep prepping

All company interviews