← Snap Interview InsightsMachine Learning Engineer

Snap·Machine Learning Engineer·Onsite - System Design / Architecture·Senior

SeniorPrefer not to say
Jun 2026

Summary

Snapchat ML Engineer interview focused entirely on designing a recommendation system for Lenses. The depth they expected was real, covering everything from candidate generation to serving pipelines, and the follow-ups on cold start and multi-signal objectives were where things got genuinely hard.

Questions Asked (3)

Q1

Design a recommendation system for Snapchat Lenses, covering problem framing, candidate generation, ranking model architecture, feature design, label/objective definition, and training and serving pipelines.

System DesignTechnical Trade-offsProduct Sense & Ideation
Author's notes

This is a big open-ended question and I spent probably too long on problem framing before they nudged me toward the actual system.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the product goal and constraints (e.g., AR lens discovery, engagement metrics, latency). Then walk through the ML system design in a structured way: problem framing, candidate generation, ranking, features, labels, and pipelines. Emphasize trade-offs and how you would measure success.

Pro tip: Snapchat's AR lenses are highly visual and ephemeral; focus on real-time signals (e.g., lens usage duration, replays, shares) and consider cold-start for new lenses. Mention how you'd handle the unique constraints of mobile AR (latency, battery, on-device vs. server).

1. Problem Framing

Define the recommendation task: what are we recommending (lenses), to whom (Snapchatters), and what business/product metrics matter (e.g., lens engagement, retention, creator ecosystem). Clarify constraints like latency, device capabilities, and privacy.

2. Candidate Generation

Design multiple candidate sources: trending/popular lenses, personalized based on user history, social signals (friends' usage), contextual (time, location, camera context), and fresh content. Use efficient retrieval methods (e.g., ANN, inverted indices) to narrow millions to hundreds.

3. Ranking Model Architecture

Propose a multi-stage ranking system: a lightweight ranker for initial filtering, then a heavier model (e.g., deep neural network with embeddings) for final ranking. Consider multi-task learning to predict multiple engagement signals (e.g., apply, share, replay).

4. Feature Design & Label Definition

List key features: user (demographics, past lens usage, friends), lens (visual features, creator, popularity, freshness), context (time, location, device). Define labels/objective: e.g., binary engagement (apply lens), or weighted multi-task (apply, share, replay) with a combined loss.

5. Training & Serving Pipelines

Describe offline training (batch, using historical logs, negative sampling) and online serving (real-time inference with low latency, possibly on-device or edge). Discuss continuous training, A/B testing, and monitoring for drift.

Key Points to Mention

  • Multi-stage ranking (candidate generation -> ranking) to balance efficiency and accuracy
  • Use of real-time and contextual features (e.g., time of day, location, camera context) for personalization
  • Multi-task learning to optimize for multiple engagement signals (apply, share, replay) simultaneously
  • Cold-start problem for new lenses: use content-based features and explore/exploit strategies
  • Latency and resource constraints: on-device vs. server inference, model compression, caching
  • Evaluation metrics: offline (AUC, NDCG) and online (CTR, engagement lift, retention) with A/B testing

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

Q2

How would you handle cold start for brand-new lenses with no engagement history, and separately for new users with no behavioral history?

System DesignTechnical Trade-offsAdaptability & Ambiguity
Author's notes

Two-part question and I made the mistake of blending them together at first.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the two cold-start problems: new lenses (items) and new users (users). Then propose a two-pronged strategy: for lenses, leverage content-based features and multi-modal embeddings; for users, use demographic/contextual signals and exploration. Emphasize evaluation and fallback mechanisms.

Pro tip: Show awareness of Snapchat's unique context: lenses are visual AR experiences, so use image/video embeddings and creator metadata; for users, leverage device, time, and location signals, but be mindful of privacy.

1. Clarify the problem and constraints

Define what 'cold start' means for lenses and users, and discuss data availability, latency, and privacy constraints at Snapchat.

2. Cold-start for new lenses

Use content-based filtering with multi-modal features (e.g., visual embeddings, creator tags, category) and possibly a bandit approach to gather initial engagement.

3. Cold-start for new users

Leverage contextual and demographic signals (e.g., device, location, time) and use popularity-based or exploration-based recommendations to bootstrap.

4. Hybrid and transfer learning

Consider meta-learning or transfer from similar users/lenses, and combine content-based and collaborative signals as data accumulates.

5. Evaluation and iteration

Define offline and online metrics (e.g., CTR, engagement, retention) and set up A/B tests to measure cold-start strategies, with fallbacks.

Key Points to Mention

  • Content-based filtering using multi-modal embeddings (image, text, audio) for lenses
  • Exploration vs. exploitation trade-off (e.g., epsilon-greedy, Thompson sampling) for new users
  • Use of contextual bandits to personalize without historical data
  • Leveraging side information: creator metadata, lens category, user demographics, device, time, location
  • Transfer learning or meta-learning from warm users/lenses
  • Evaluation metrics: CTR, engagement, retention, and A/B testing framework

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

Q3

Users interact with lenses at different levels: viewing, trying, sharing, saving. How would you model these multiple engagement signals? Walk through the trade-offs between a single model with weighted loss terms versus separate models per action combined into a final score.

Technical Trade-offsSystem DesignProduct Analytics & Metrics
Author's notes

This was the follow-up I felt best about.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by framing the problem as a multi-task learning challenge with a shared user-item representation, then compare the two modeling strategies across data efficiency, flexibility, and production complexity. Conclude with a recommendation that balances business goals (e.g., predicting overall engagement) with engineering constraints (e.g., latency, maintenance).

Pro tip: Emphasize that the choice depends on whether the engagement signals are complementary or competing; if they are complementary, a multi-task model with learned weights often outperforms separate models, but if they are competing, separate models with a calibrated fusion layer may be better. Also, mention the importance of monitoring for negative transfer and using techniques like GradNorm or uncertainty weighting.

1. Define the objective and signals

Clarify what the final goal is (e.g., predict overall engagement or rank lenses) and characterize each signal (view, try, share, save) in terms of frequency, sparsity, and business value.

2. Single model with weighted loss

Describe a multi-task model with a shared bottom and task-specific heads, where the total loss is a weighted sum of per-task losses. Discuss how to set weights (manual, uncertainty weighting, GradNorm) and the benefits (data efficiency, shared representations) and drawbacks (negative transfer, weight tuning).

3. Separate models per action

Describe training independent models for each action and combining their outputs into a final score via a simple rule (e.g., weighted sum) or a learned fusion model. Discuss benefits (flexibility, no interference) and drawbacks (more data/compute, no shared learning, fusion complexity).

4. Compare trade-offs

Contrast the two approaches on dimensions like data efficiency, model complexity, inference latency, ease of updating individual tasks, and ability to handle task correlations.

5. Recommend and justify

Propose a solution based on the trade-offs, considering Snapchat's scale and product needs. For example, start with a multi-task model for efficiency, but if tasks conflict, switch to separate models with a learned fusion layer.

Key Points to Mention

  • Multi-task learning with shared representations and task-specific heads
  • Loss weighting strategies: uncertainty weighting, GradNorm, dynamic weight averaging
  • Negative transfer and how to mitigate it (e.g., task grouping, adversarial training)
  • Separate models: independence, ease of debugging, but higher maintenance and no shared learning
  • Fusion methods for separate models: linear combination, stacking, or a small neural network
  • Evaluation metrics: per-task AUC, overall engagement metric (e.g., NDCG), and online A/B testing

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