Build a scalable A/B Testing Pipeline
Last updated: January 14, 2026
Quick Overview
Design a scalable a/b testing system that handles millions of requests. Discuss trade-offs in consistency, availability, and performance.
Two Sigma
January 14, 202637
4
2,853 solved
Design a scalable a/b testing system that handles millions of requests. Discuss trade-offs in consistency, availability, and performance.
ML system design at Two Sigma goes beyond model selection. This System Design Round question evaluates your ability to design end-to-end ML pipelines, from data collection to model serving, while considering production constraints like latency and reliability.
What the Interviewer Expects
- Define clear ML objectives with appropriate loss functions and metrics
- Design a comprehensive feature engineering pipeline
- Discuss model selection with trade-offs (complexity vs interpretability vs latency)
- Plan online and offline evaluation strategies including A/B testing
- Address serving infrastructure: batch vs real-time, latency requirements
- Consider data quality, labeling strategy, and feedback loops
Key Topics to Cover
How to Approach This
- Start by clarifying functional and non-functional requirements with the interviewer.
- Estimate the scale: QPS, storage, bandwidth. This drives your design decisions.
- Draw a high-level architecture first, then deep dive into 1-2 critical components.
- Discuss trade-offs explicitly (e.g., consistency vs availability, SQL vs NoSQL).
- Address failure scenarios, monitoring, and how the system handles 10x traffic spikes.
Possible Follow-up Questions
- How would you handle the cold start problem?
- How would you ensure fairness and reduce bias in the model?
- What is your model retraining strategy?
Practice a Similar Problem on Codemia
Solve a related problem with our interactive workspace, get AI feedback, and view detailed solutions.
Solve on CodemiaSample Answer
Requirements
- Functional Requirements:
- Ability to create, manage, and analyze A/B tests for multiple ML models.
- Real-time data collection and user assignment to control or variant groups.
- Dashboar...
Capacity Estimation
Assuming we handle 10 million requests per day:
-
Requests per second (RPS):
- 10,000,000 requests/day = ~115 RPS.
-
User Assignments:
- Each A/B test might involve 2 groups (A and B). ...