Design a Product Tagging Pipeline
Company: Snapchat
Role: Machine Learning Engineer
Category: ML System Design
Difficulty: medium
Interview Round: Onsite
Design an applied machine learning pipeline that automatically assigns standardized tags to products and seller content in a short-video commerce platform. The tags are used for search, recommendations, moderation, and analytics.
Discuss:
1. The tag taxonomy and scope.
2. Data sources and label generation.
3. Modeling choices for text, image, and video signals.
4. Offline training and online inference.
5. Cold start, low-confidence predictions, and human review.
6. Evaluation, monitoring, and feedback loops.
Quick Answer: This question evaluates applied machine learning system design, multi-modal modeling, data engineering, and production deployment competencies involved in automatically assigning standardized tags to product and seller content across text, image, and video signals.