This part preserves 11 Microsoft AI/ML internship reports from 2020.

Coverage: 2020 · candidate reports 108–118 · phone screens and final loops

Series note: Part 1 focuses on the CoreAI / AI-ML Software Engineering Intern loop. This part preserves other Microsoft AI/ML internship reports without treating them as CoreAI SWE evidence.

2020 reports (continued)

108: Data Scientist Intern, A/B testing team — final interview loop

Report date: 2020-02-21 · Role: Data Scientist Intern, A/B testing team · Stage: Final loop / onsite · Interview date: 2020-02-11

Microsoft Data Scientist internship onsite for an A/B-testing team. The four rounds covered behavioral and basic ML questions, an undisclosed A/B-testing case, an unnamed easy-to-medium LeetCode-style coding task, and probability/statistics including a vaguely recalled dice-expectation item.

1. No standalone technical prompt was preserved. This entry is retained for its process, timing, role, or interview-format evidence.

Other reported topics or limits: Undisclosed A/B-Testing Case and Coding Question; Vague Dice-Expectation and Random-Number Questions; Behavioral and Basic ML Questions.

109: Data Science PhD Intern — final interview loop

Report date: 2020-02-13 · Role: Data Science PhD Intern · Stage: Final loop / onsite · Interview date: 2020-02-11

Detailed Microsoft Data Science PhD internship virtual onsite across Bing Ads and a market/Excel team.

1. Number of Islands. The coding task was Number of Islands.
2. Design Semantic Ad Retrieval for a Query. Given a query word such as 'shoes', retrieve relevant content from all ads. The discussion considered knowledge-graph word expansion, word embeddings, and learned similarity between the query and ad text or images.
3. Design an Excel Next-Command Recommender. Recommend which Excel command a user is likely to use next based on actions in the current workbook/session. Available fields include time, user ID, Excel session ID, and command. Explain how to evaluate the recommendation system.

Other reported topics or limits: ML Concept and Resume Deep-Dive Questions.

110: Microsoft Research NLP Intern — phone or technical screen

Report date: 2020-02-03 · Role: Microsoft Research NLP Intern · Stage: Phone / technical screen

Detailed Microsoft Research NLP internship report.

1. Return the Top K Search-Query Suggestions for a Prefix. A collection contains many search queries. Given the first several words of a string/query, return the top k likely suggested queries. The author identified a trie as the intended data structure.
2. Implement a Direction-Specific Research Model. The interviewer asked whether the candidate knew a model related to the target internship direction and then asked them to implement it. The report did not preserve a reconstructable task contract.
3. Implement a Sequence-to-Sequence Model. The second interviewer asked the candidate to implement seq2seq. The candidate wrote a short implementation and corrected one pointed-out mistake. The report did not preserve a reconstructable task contract.

Other reported topics or limits: Paper and Research Discussion.

111: Data & Applied Science Intern — phone or technical screen

Report date: 2020-01-30 · Role: Data & Applied Science Intern · Stage: Phone / technical screen · Interview date: 2020-01-22

Detailed Microsoft Data & Applied Science internship phone screen. The discussion focused on commonly used ML/statistical methods, one relevant project in depth, project challenges, explaining deep learning, and improving a favorite Microsoft product; no coding was asked.

1. No standalone technical prompt was preserved. This entry is retained for its process, timing, role, or interview-format evidence.

Other reported topics or limits: ML Method, Project Deep-Dive, and Product-Improvement Screen.

112: Data Scientist PhD Intern, Bing team — final interview loop

Report date: 2020-01-28 · Role: Data Scientist PhD Intern, Bing team · Stage: Final loop / onsite · Interview date: 2020-01-16

Detailed Microsoft Data Scientist PhD internship process for a Bing/NLP-ML team.

1. Knapsack Problem Variants. The interview included several versions of the knapsack problem, with dynamic programming as the focus.
2. Largest K in an Array Using a Heap. Find the largest K values in an array using a heap.

Other reported topics or limits: Unspecified NLP List-Processing Task; ML and DL Concept-Comparison Questions; Recruiter Behavioral and Product Screen.

113: Data Analysis / Machine Learning Intern — final interview loop

Report date: 2020-01-26 · Role: Data Analysis / Machine Learning Intern · Stage: Final loop / onsite

Microsoft Japan summer internship onsite report for an input-method team seeking data-analysis help.

1. Print Vertical Asterisk Bars. The input is a sequence of numbers representing bar heights. The requested output draws the bars with asterisks. For heights 3, 1, 2, the example uses three rows with bottom-aligned vertical bars.
2. Word Search with Eight-Direction Adjacency. The input contains an N by N matrix filled with characters and a target string. The task is to determine whether the target string exists in the matrix. Horizontal, vertical, and diagonal moves are all valid, and a brute-force solution was acceptable.

Other reported topics or limits: Score Comparison and ML Discussion.

114: Data and Applied Scientist Intern — phone or technical screen

Report date: 2020-01-24 · Role: Data and Applied Scientist Intern · Stage: Phone / technical screen · Interview date: 2020-01-24

Detailed Microsoft Data and Applied Scientist internship recruiter screen. The interviewer asked the candidate to choose an ML/DL domain, then covered project challenge and validation, explaining deep learning to a child, language preferences, end-to-end ML workflow, motivation, and product improvement. No coding was asked.

1. No standalone technical prompt was preserved. This entry is retained for its process, timing, role, or interview-format evidence.

Other reported topics or limits: Recruiter ML Background, Workflow, and Product Screen.

115: Data Scientist PhD Intern — phone or technical screen

Report date: 2020-01-23 · Role: Data Scientist PhD Intern · Stage: Phone / technical screen

Microsoft Data Scientist PhD internship phone screen. After a project and p-value discussion, the interviewer asked one fully described probability question about repeated rolls of two dice and which target sum appears first.

1. No standalone technical prompt was preserved. This entry is retained for its process, timing, role, or interview-format evidence.

Other reported topics or limits: Probability That Sum Seven Appears Before Sum Eight; Project and P-Value Discussion.

116: Data and Applied Scientist Intern — phone or technical screen

Report date: 2020-01-22 · Role: Data and Applied Scientist Intern · Stage: Phone / technical screen

Microsoft Data and Applied Scientist internship first-round recruiter screen. The questions covered graduation timing, DS versus ML preference, product improvement, a challenging project and its data/language details, difficult teammate handling, prior ML tools and algorithms, and company knowledge.

1. No standalone technical prompt was preserved. This entry is retained for its process, timing, role, or interview-format evidence.

Other reported topics or limits: Recruiter Background, Project, Product, and Teamwork Screen.

117: Data Scientist Intern — phone or technical screen

Report date: 2020-01-17 · Role: Data Scientist Intern · Stage: Phone / technical screen

Detailed Microsoft Data Scientist internship technical phone screen. The interviewer conducted a sustained linear-regression and estimation theory discussion covering coefficient interpretation, normalization, predictor choice, nonlinearity, interactions, signs, least squares, MLE, and their equivalence.

1. No standalone technical prompt was preserved. This entry is retained for its process, timing, role, or interview-format evidence.

Other reported topics or limits: Linear Regression and Estimation-Theory Discussion; Graduation Date and Project Deep Dive.

118: Data & Applied Science PhD Intern — phone or technical screen

Report date: 2020-01-14 · Role: Data & Applied Science PhD Intern · Stage: Phone / technical screen · Interview date: 2020-01-14

Microsoft Data & Applied Science PhD internship technical phone screen. The discussion covered LSTM choice and metrics, gradient descent, poor validation performance despite strong training performance, and L1 versus L2 regularization.

1. No standalone technical prompt was preserved. This entry is retained for its process, timing, role, or interview-format evidence.

Other reported topics or limits: LSTM, Gradient Descent, Generalization, and Regularization Discussion; Background, Experience, and Future-Project Discussion.

What repeated often enough to prioritize

Tier 1: Practice until automatic

1. Arrays, strings, hash maps, heaps, BFS/DFS, and clean complexity analysis.
2. One end-to-end ML project explanation: data, labels, features, model choice, validation, failure analysis, and deployment trade-offs.
3. Precision, recall, confusion matrices, threshold trade-offs, and choosing metrics for a product objective.

Tier 2: Be ready to explain deeply

1. Experiment design when ordinary A/B testing is unavailable or biased.
2. Recommendation, ranking, retrieval, and classification systems, including feedback loops and label quality.
3. Probability and sampling questions where assumptions, fairness, and simulation behavior matter.

Tier 3: Higher-level differentiators

1. Research depth: why a method was chosen and what evidence would change that choice.
2. Product diagnosis under sudden metric changes, sparse feedback, drift, or imperfect ground truth.
3. Clear communication when a prompt is underspecified: ask for the missing contract instead of silently inventing it.

Final checklist

Coding

☐ Rehearse arrays, strings, linked lists, trees, graphs, heaps, dynamic programming, and binary search.

☐ State the input contract, edge cases, complexity, and optimization path before polishing code.

☐ Practice the linked FastPrep questions above in Java, Python, C++, or Go.

ML and statistics

☐ Explain precision, recall, threshold trade-offs, bias/variance, regularization, loss functions, and validation failure modes in plain language.

☐ Work through sampling, probability, estimation, and hypothesis-testing cases without using a library call as the explanation.

ML/product design

☐ Practice label design, offline and online metrics, experiment constraints, feedback loops, drift, and launch monitoring.

☐ Prepare one retrieval/ranking/recommendation design and one product-metric diagnosis end to end.

Behavioral and research depth

☐ Prepare concise stories for conflict, influence, failure, ambiguity, and why Microsoft or the target team.

☐ Defend every line of your strongest project: your contribution, alternatives, and what failed.

This draft covers candidate reports from 2020 through 2026. Duplicate and status-only follow-ups were merged or excluded. Missing prompt mechanics remain marked as missing. Corrections and richer reports are welcome in a future revision.

Part 10 keeps the remaining secondhand, duplicate, status-only, pre-interview, and adjacent-role context without adding those records to the 118-experience count.