Data Engineering, ML Foundations, and Deep Learning Essentials
Free GSDC Certified Forward Deployed Engineer practice — 6 questions on Data Engineering, ML Foundations, and Deep Learning Essentials, with explanations. No sign-up. Full 12-question mixed test →
A fintech company processes credit card transactions and must flag potential fraud within 200ms of each transaction occurring, using continuously updated behavioral features. Which data pipeline pattern BEST meets this requirement?
A streaming pipeline processes events as they arrive and can score fraud within milliseconds, meeting the 200ms latency requirement with continuously updated features.
A team building a RAG chatbot needs to store 2 million document chunks and retrieve the most semantically similar chunks to a user's query in milliseconds. Which storage solution is the BEST fit?
Vector databases index embeddings using ANN algorithms (e.g., HNSW), enabling fast semantic similarity search at millions-of-vectors scale, which is exactly what RAG retrieval requires.
You are tuning a binary classifier for a highly imbalanced dataset (2% positive class) using k-fold cross-validation. Which configuration choice is MOST important to avoid biased performance estimates?
Stratified k-fold ensures every fold maintains the same 2% positive-class ratio as the full dataset, preventing folds that contain zero or very few positive examples and producing more reliable evaluation metrics.
What is the primary advantage of the self-attention mechanism in transformer architectures compared to the sequential processing in RNNs?
Self-attention computes weighted relationships between every pair of tokens simultaneously, enabling parallel computation (unlike RNNs' step-by-step processing) and directly modeling long-range dependencies without vanishing-gradient issues.
A data scientist trained a model last quarter that performed well, but the team can no longer reproduce those exact results despite having the saved model file. Which combination of logged artifacts would have MOST reliably ensured reproducibility?
True reproducibility requires capturing the exact code version, dataset snapshot, hyperparameters, and environment/library versions — this is precisely what experiment tracking tools like MLflow or Weights & Biases are designed to log automatically.
A company wants to automatically route incoming support tickets into exactly 5 fixed categories based on well-defined keyword rules (e.g., the word 'refund' in a ticket always maps to 'Billing'), and these rules rarely change. Which approach is MOST appropriate?
When categories are few, fixed, and mapped by clear deterministic keyword rules that rarely change, a rule-based system is the simplest, cheapest, most maintainable, and fully explainable solution — exactly the scenario where the exam expects you to avoid over-engineering with ML/LLM.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$109.99 $34.99 with code FREETEST33 — valid through September 14.
Get my $34.99 deal →