← Back to Directory

Microsoft interview notes, page 17 of 60

FAANG / Big Tech

Most of these Microsoft write-ups come from Software Engineers (372), then Product Managers (124) and Machine Learning Engineers (53). The most reported rounds are the technical phone screen (255), the product sense round (111), the system design round (93) and the behavioral / leadership round (49).

597 interview notes · updated Oct 2026

Microsoft·Software Engineer·Onsite - System Design / Architecture

May 2026
Microsoft system design round, one question about building a URL shortener from scratch. Pretty open-ended and they clearly wanted to see how far you'd push the architecture before they had to prompt you.
  • Design and architect a link shortening service similar to Bit.ly. Walk through the full system.

“I started with the basic encode/decode flow and felt pretty good about that part.”

View Post

Microsoft·Software Engineer·Onsite - Coding / Algorithms

May 2026
Third round at Microsoft for a software engineering role. It was a LeetCode-style coding problem, a variant of a well-known medium/hard problem, with the input format changed to use two string arrays instead of the original. Had to run tests before wrapping up.
  • Solve a variant of a well-known LeetCode problem (700s difficulty range) where the input is modified to accept two string arrays instead of the original input type.

“The core logic wasn't totally foreign but the input change threw me off more than I expected.”

View Post

Microsoft·Machine Learning Engineer·Onsite - System Design / Architecture

May 2026
System design round at Microsoft for an MLE role, focused entirely on building a product search system for a large e-commerce platform. The question was broad and covered a lot of ground, from indexing pipelines to ranking models to serving architecture. Pretty intense for a single session.
  • Design a product search system for a large e-commerce marketplace that supports free-text queries, filters, typo tolerance, synonyms, personalization, and relevance ranking.
  • Walk through the query-time flow from when a user submits a search request to when they see a ranked list of results.
  • How would you design the product ingestion, indexing, and update pipeline to handle frequent catalog changes like price and inventory updates?
  • What retrieval strategy would you use, and how would you combine lexical and semantic retrieval?
  • How would you design the ranking model and generate features for it?
  • How would you evaluate the search system both offline and online?
  • What trade-offs would you make between latency, freshness, relevance, and cost in this system?

“This question is basically a full ML system design in one shot.”

View Post

Microsoft·Software Engineer·Onsite - System Design / Architecture

May 2026
Microsoft system design round, one big question about building a distributed key-value store with replication. Pretty intense for a single question but they clearly wanted to see how deep you could go on consistency and fault tolerance.
  • Design a distributed key-value store that replicates data across multiple nodes, handles quorum-based reads and writes, maintains consistency, and degrades gracefully when nodes go down.

“I went straight to the replication factor and quorum math (W + R > N) which was the right call, but I fumbled when they pushed on what happens during a network partition.”

View Post

Microsoft·Software Engineer·Technical Phone Screen

May 2026
Microsoft software engineering interview with a data-heavy problem that had a few layers to it. Not a pure coding question, more of a design-and-reasoning exercise that kept expanding the more you answered.
  • You have a dataset of concert venue vote records. How do you find the most popular city, handle inconsistent city name spellings like 'NYC' vs 'New York', deal with votes that have different weights, and scale the whole thing for large datasets?

“This question kept growing.”

View Post

Microsoft·Data Scientist·Technical Phone Screen

May 2026
Got a machine learning theory question at Microsoft, just the one. Short and focused, felt more like a screen than a deep technical dive.
  • Is Mean Squared Error (MSE) an appropriate cost function to use when training a logistic regression model?

“Said no pretty quickly, which was right, but then stumbled trying to explain why in a clean way.”

View Post

Microsoft·Data Scientist·Technical Phone Screen

May 2026
Phone screen with Microsoft for a data scientist role, mostly focused on ML fundamentals and how I talk about my own work. Pretty conversational but they covered a lot of ground fast.
  • What kinds of data do you typically work with in your projects?
  • Walk me through a machine learning pipeline you've built end to end.
  • How would you explain deep learning to a five-year-old?
  • Which deep learning model architectures have you actually used, and what were the use cases?

“Easy opener but I rambled.”

View Post

Microsoft·Software Engineer·Technical Phone Screen

May 2026
Microsoft SWE interview, just one question about multithreading. Not much to go on but here's what I remember.
  • Can you explain multithreading?

“Pretty open-ended.”

View Post

Microsoft·Software Engineer·Technical Phone Screen

May 2026
Microsoft SWE interview with a data aggregation problem that looked straightforward but had enough edge cases to keep you busy for a while.
  • Given a dataset of concert location voting records (each with a voter ID, a location string, and an optional numeric weight defaulting to 1), write SQL and/or Python to find the most popular location by total vote weight. Cover how you handle ties, missing or malformed location text, and walk through a small example to verify. Also analyze time and space complexity.

“I jumped straight into the SQL groupby-sum approach and felt pretty good about it, then they asked about ties and I kind of stumbled.”

View Post

Microsoft·Software Engineer·Technical Phone Screen

May 2026
Microsoft Applied Scientist interview focused heavily on LLM fundamentals, covering both the theoretical side of transformer architecture and the messier real-world stuff like latency, cost, and safety. Two broad areas but a lot of ground to cover in each.
  • Walk me through how self-attention works in a transformer, including multi-head attention and positional encodings.
  • What are the differences between encoder-only, decoder-only, and encoder-decoder transformer architectures, and when would you use each?
  • Explain what a KV cache is and why it matters for inference.
  • How would you address hallucination in a production LLM system?
  • What techniques would you use to reduce inference latency for a large language model?
  • How do continuous batching and paged attention help with the cost of serving LLMs?
  • How would you approach safety and content moderation when deploying an LLM in a real product?
  • How do tensor parallelism and pipeline parallelism differ, and when would you use one over the other?

“Felt pretty solid on self-attention mechanics but stumbled a bit explaining why you'd want multiple heads instead of just one big attention matrix.”

View Post