Four firsthand reports that help answer two questions: Is the loop always three back-to-back interviews, and is it mostly LeetCode or does it include ML?

Coverage: Microsoft CoreAI and closely related AI/ML Software Engineering Intern roles · 2025–2026

The short answer

There is no single interview loop that every candidate gets.

1. Three back-to-back interviews are common, but not guaranteed. Two candidates described three interviews. One CoreAI Business candidate had two. Another CoreAI candidate listed three parts of the loop but did not share the exact order or timing.
2. Coding is the safest thing to expect. All four reports included coding. The questions ranged from LeetCode Easy to Medium.
3. ML can show up, but it may be light. One candidate got two basic ML questions. Another had a general AI/ML discussion. A third reported three coding rounds and did not mention ML at all.
4. Some newer loops also include system design or AI-assisted coding. These showed up in two 2026 CoreAI reports.

So if your recruiter told you there will be three 45-minute interviews, do not assume all three will be LeetCode. Prepare for coding first, but also be ready to discuss your resume, basic ML, and possibly system design.

What the four candidates reported

01: Three 45-minute interviews with coding, behavioral, and ML

Interview date: September 18, 2025 · Role: AI/ML Software Engineering Intern · Team: CoreAI

This candidate had three 45-minute one-on-one interviews, with a 15-minute break between each one.

Round 1

• Behavioral questions
• A coding problem about checking whether two strings have a one-to-one character mapping
• The candidate wrote the solution quickly, but the code did not produce output before time ran out

You can practice the same coding idea with Isomorphic Strings.

Round 2

• Behavioral discussion
• One LeetCode Easy problem and a follow-up; the exact question was not shared
• Two ML questions

Only one ML question was described. The interviewer asked how to represent Bachelor's and Master's degrees in a linear regression model, then asked what should change if PhD is added as a third category.

Round 3

• Behavioral questions
• Merge two linked lists that are already sorted in descending order
• Implement a linked list; the exact operations were not shared
• Sort one randomly ordered linked list in descending order by reusing the earlier merge logic
• Explain time and space complexity and walk through test cases

Related practice:

• Merge Two Descending Linked Lists
• Sort a Linked List in Descending Order

This report is the clearest match for the three-round question. It was mostly coding, but it was not all LeetCode. ML and behavioral questions were part of the same loop.

02: Three coding-heavy interviews with no ML questions reported

Report date: November 27, 2025 · Role: AI/ML Software Engineering Intern

This candidate also had three interviews.

• Each interview included one LeetCode Medium problem.
• The interviewers also asked about the candidate's resume and behavioral examples.
• The candidate did not report any ML or system-design questions.
• The exact coding problems were not shared.

This is a good reminder that an AI/ML job title does not always mean there will be a separate ML round. Some candidates still get a mostly standard SWE loop.

03: One coding interview and one system-design interview

Interview date: February 6, 2026 · Role: AI/ML Software Engineering Intern · Team: CoreAI Business

This candidate had two interviews, not three.

Interview 1

• One LeetCode Easy problem
• A deep discussion about the candidate's resume
• A general conversation about AI and ML

Interview 2

• System design
• The discussion focused on databases and systems

The candidate did not share the exact AI/ML questions or the system-design prompt. This loop mixed coding, ML conversation, and system design, but it did not use the three-interview format.

04: Traditional coding, AI-assisted coding, and system design

Report date: September 24, 2026 · Role: Software Engineering Intern · Team: CoreAI

This candidate described three parts of the interview loop:

• Traditional coding
• AI-assisted coding
• System design

The candidate did not share the questions, the exact order, or the length of each interview. They also said the questions could vary a lot between interviewers.

This is the newest report in the group. It suggests that some CoreAI loops now test how candidates work with AI tools, not only how they solve a normal coding problem.

What to prepare

Coding

1. Practice LeetCode Easy and Medium problems under a 45-minute clock.
2. Be comfortable with strings, hash maps, and linked lists.
3. Explain your approach, complexity, edge cases, and test cases while you code.

ML

1. Review basic ML concepts connected to your own projects.
2. Know how categorical features are represented, including one-hot or dummy encoding.
3. Be ready to explain your data, model choice, evaluation method, and tradeoffs in plain English.

Other parts of the loop

1. Prepare short behavioral stories and know your resume well.
2. Practice one intern-level system-design interview.
3. Try one AI-assisted coding mock where you check the generated code, test it, and explain every change you keep.