Product Context
CareerMatch is a professional jobs platform. You are leading the team building a personalized job recommendation system for the home feed and email digests, helping job seekers discover relevant openings while helping employers receive qualified applicants.
Scale
| Signal | Value |
|---|
| DAU | 18M job seekers |
| Peak recommendation QPS | 45K |
| Active job catalog | 12M open jobs |
| New / updated jobs per day | 1.1M |
| Applications per day | 9M |
| End-to-end p99 latency budget | 180ms |
Task
You are the technical lead responsible for the system design. Walk through how you would design and deliver this ML solution end to end.
- Clarify the product goal, target users, and success metrics for recommendations
- Propose a multi-stage architecture for candidate generation, ranking, and re-ranking
- Define the offline and online data pipelines, including labels, features, and retraining cadence
- Design the serving architecture, including online vs batch recommendations, caching, and fallback behavior
- Explain how you would evaluate the system offline and online, then launch it safely
- Identify likely failure modes, especially feature drift, training-serving skew, and cold start, and explain mitigations
Constraints
- Users expect fresh recommendations when jobs are posted or updated; new jobs should be eligible within 10 minutes
- Many users are sparse-history or first-time visitors, so cold start is material
- Some recommendations are delivered online in-session, while daily email digests can be precomputed in batch
- The system must support policy filters such as location, visa eligibility, blocked employers, and seniority constraints
- Cost matters: the online serving path should avoid expensive per-request deep models on the full catalog
- Compliance requirement: retain feature snapshots for auditability of recommendation decisions
