Dropbox·Machine Learning Engineer·Technical Phone Screen·Senior

SeniorPrefer not to say

May 2026

Summary

ML engineer design question at Dropbox, basically one big open-ended problem about building a file recommendation feature for the mobile app. No behavioral stuff from what I remember, just deep on the system.

Questions Asked (1)

Design a recommender system that surfaces relevant files to users when they open the Dropbox mobile app.

System DesignProduct Sense & IdeationTechnical Trade-offs

Author's notes

I jumped straight into collaborative filtering and the interviewer kind of just waited.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the product goal and constraints (e.g., latency, privacy, cold-start) and then propose a two-stage retrieval-ranking architecture. Emphasize how you would leverage Dropbox-specific signals like file access patterns, sharing activity, and content embeddings to personalize recommendations.

Pro tip: Show awareness of the mobile constraints by discussing on-device vs. server-side trade-offs and how you'd handle privacy-sensitive data. Also, mention how you'd measure success beyond clicks, such as file open rate or task completion.

1. Clarify requirements and constraints

Ask questions to understand the goal: what defines a 'relevant' file? What are latency, privacy, and scale constraints? How does the mobile context differ from desktop?

2. Define success metrics and data sources

Identify offline and online metrics (e.g., recall@k, file open rate) and list available signals: user behavior, file metadata, content, and social/sharing graph.

3. Propose a high-level architecture

Outline a two-stage system: candidate generation (e.g., collaborative filtering, content-based) followed by ranking (e.g., gradient boosted trees or neural network) with business rules.

4. Detail model choices and features

Explain specific models (e.g., matrix factorization for retrieval, DNN for ranking) and key features (recency, frequency, file type, collaborators, embeddings).

5. Address trade-offs and deployment

Discuss trade-offs (e.g., latency vs. accuracy, on-device vs. cloud), cold-start solutions, and how to deploy and monitor the system on mobile.

Key Points to Mention

Two-stage retrieval-ranking architecture to balance scalability and accuracy
Use of Dropbox-specific signals: file access patterns, sharing activity, collaborators, and content embeddings
Handling cold-start for new users and new files via content-based and heuristic methods
Mobile constraints: latency, bandwidth, battery, and on-device inference options
Privacy and security considerations, including on-device processing and differential privacy
Evaluation strategy: offline metrics, online A/B testing, and long-term user engagement

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.