Where should you store features for low-latency online prediction and point-in-time training?
You work for a delivery company. You need to design a system that stores and manages features such as parcels delivered and truck locations over time. The system must retrieve the features with low latency and feed those features into a model for online prediction. The data science team will retrieve historical data at a specific point in time for model training. You want to store the features with minimal effort. What should you do?
Community Votes
100% of anonymous learners picked answer B. Votes are pick records left by other test-takers — they are not the verified answer.
Community Insight
The question tests whether you can map three trigger phrases — low latency, point-in-time historical lookup, and minimal effort — to Vertex AI Feature Store instead of a raw storage product.
Vertex AI Feature Store is Google Cloud's managed solution for serving features with low latency to online predictions while supporting point-in-time retrieval of historical data for training. The PMLE community votes 100% for Feature Store, citing its low-latency serving, time-travel lookups, and minimal operational effort.
The most tempting wrong answer is A (Bigtable), because it also offers low-latency key/value reads; however, it forces you to design schemas, implement feature versioning, and build custom point-in-time logic yourself, which violates the minimal-effort requirement.
Community Discussion (6 comments)
Comments & Corrections
No comments yet — spotted an error or have a note? Share it below.
Expert Analysis
Why the Answer Is Correct
Vertex AI Feature Store is purpose-built for exactly this scenario. It provides a managed online serving layer that returns feature values with low latency for online prediction, and it supports point-in-time (time travel) lookups so the data science team can fetch feature values as of a specific timestamp for training. Because it is fully managed, there is no schema engineering, no custom serving layer, and no manual synchronization between offline and online stores — satisfying the minimal-effort requirement. Community comment [1] explicitly maps the three question keywords (low latency, point-in-time retrieval, minimal effort) to Feature Store capabilities.Why the Other Options Are Wrong
Option A (Bigtable) delivers low-latency key/value access, but you would have to design the row-key schema, implement feature versioning, and build your own point-in-time retrieval and serving logic — significant custom effort. Option C (Vertex AI dataset) is merely a container for training data and cannot serve features to online predictions at low latency. Option D (BigQuery timestamp-partitioned tables with the Storage Read API) is strong for analytics and historical batch reads, but the Storage Read API is a batch-oriented pipeline, not an ultra-low-latency online feature-serving system.Community Comment Notes
The vote distribution is 100% for B, making this one of the most consensual PMLE questions. Comment [1] methodically quotes the three requirement phrases from the question and ties each to a Feature Store capability — a verification technique worth copying on exam day. Comment [3] adds that Feature Store is optimized for ultra-low latency serving, and comment [4] confirms its point-in-time retrieval support. There are no dissenting answers or contested debate in the discussion.Official Reference
Exam Strategy
On the PMLE exam, whenever a question mentions serving features for online prediction, check Vertex AI Feature Store first. Anchor on the trigger phrases 'low latency,' 'point-in-time lookup,' and 'minimal effort,' and eliminate storage-only options like Bigtable or BigQuery that would require custom engineering to become a feature store.
Related Analysis
Practice All PMLE Questions
Access 65 questions with complete answers and detailed explanations.
View Full PMLE Practice Test →