Design Comprehensive Recommendation System for Spokeo Features
Overview
This question evaluates the ability to design an end-to-end recommendation system for a people-search platform, testing competencies in data collection (batch and streaming), feature engineering, candidate generation and ranking, offline and online evaluation metrics, real-time serving architecture, experimentation, and cold-start handling.
Spokeo
Aug 4, 2025, 10:55 AM
Data Scientist
Onsite
Machine Learning
0
0
Design an End-to-End Recommendation System for Spokeo
Scenario
You are designing a new recommendation system for Spokeo (a people-search platform) to help users find relevant profiles and related searches more quickly and safely.
Task
Propose an end-to-end design that covers:
- Data collection and pipeline (batch + streaming)
- Feature engineering
- Model selection (candidate generation + ranking)
- Evaluation metrics (offline and online)
- Real-time serving architecture
- A/B testing and experimentation plan
Include how you will handle cold-start users/items, and specify both offline and online metrics.
Assumptions
- Primary recommendation surfaces: (a) search results and (b) profile detail pages.
- Target recommendations: "profiles you might be looking for" and "related searches".
- Objectives: improve relevance (find the right person), efficiency (fewer reformulations), and safety (avoid harmful/sensitive suggestions).
Solution
Login required
Show