← Intuit Interview Insights

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

Senior
Apr 2026

Summary

Interviewed for an MLE role at Intuit, second session was a case study focused on recommendation systems. The question was meaty and pretty open-ended, which I wasn't fully prepared for.

Questions Asked (1)

Q1

How would you design an offline personalized promotional product recommendation system that handles delayed and sparse conversion feedback?

System DesignTechnical Trade-offsProduct Analytics & Metrics
Author's notes

This one exposed some gaps in how I think about feedback loops.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business context and constraints, then outline a system design that addresses delayed and sparse feedback through a combination of delayed feedback modeling, exploration, and offline evaluation. Emphasize trade-offs between model complexity, latency, and business impact, and propose a phased implementation plan.

Pro tip: Show that you understand the difference between offline metrics and online business metrics, and propose a robust experimentation framework to validate the system in production. Also, highlight the importance of data leakage prevention when using delayed labels.

1. Clarify Requirements and Constraints

Ask questions to understand the business goals, data availability, latency requirements, and scale. Identify what 'promotional product' means and how conversions are defined.

2. Design Data Pipeline and Feature Store

Outline how to ingest and process user interactions, product metadata, and delayed conversion events. Propose a feature store that supports both batch and real-time features.

3. Model Delayed and Sparse Feedback

Discuss techniques like survival analysis, delayed feedback modeling, and using auxiliary signals. Address sparsity with matrix factorization, content-based methods, or graph neural networks.

4. Handle Exploration and Cold Start

Incorporate exploration strategies (e.g., Thompson sampling, epsilon-greedy) to gather feedback on new products and users. Use content-based features to mitigate cold start.

5. Offline Evaluation and Online Deployment

Propose offline evaluation methods that account for delayed feedback (e.g., inverse propensity scoring, time-based splits). Plan for A/B testing and monitoring of business metrics.

Key Points to Mention

  • Delayed feedback modeling (e.g., survival analysis, delayed feedback in bandits)
  • Sparse conversion handling (e.g., matrix factorization, content-based filtering, graph neural networks)
  • Exploration vs. exploitation trade-off (e.g., Thompson sampling, contextual bandits)
  • Offline evaluation with delayed labels (e.g., inverse propensity scoring, time-based validation)
  • Feature store and data pipeline design for batch and real-time
  • Business metrics and A/B testing for online validation

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