Covers Introduction to Neural Networks, Neurons & Perceptrons, and Activation Functions. Tests your understanding of the building blocks of any neural network — perceptron geometry, universal approximation, and activation characteristics. Ideal for warming up before deeper training-mechanics topics.

Covers Forward Propagation, Loss & Cost Functions, and Backpropagation. Tests how information flows through a network, how error is quantified, and how gradients propagate back. Critical for understanding why networks learn (or fail to).

Covers Optimizers and ANN Architectures. Tests gradient-descent variants, adaptive learning rates, momentum, and how macro-level architectural choices (depth, width, skip connections) affect training dynamics. Ideal prep for questions on training instability and scaling.

Covers Regularization & Normalization and Weight Initialization. Tests dropout mechanics, BatchNorm vs LayerNorm trade-offs, covariate shift, and initialization schemes (Xavier, He, LeCun). Understanding this cluster is essential for training stable, generalizable networks.

Covers CNN Architectures and RNN / LSTM / GRU. Tests spatial feature extraction, receptive fields, residual connections, and sequential memory mechanisms. A must-have before moving to Attention — these are the baselines Transformers replaced.

Covers Attention Mechanisms & Transformers and Self-Supervised & Contrastive Learning. Tests scaled dot-product attention, positional encodings, multi-head attention, and pretraining paradigms like SimCLR, MoCo, and MAE. Core to any modern ML interview.

Covers Graph Neural Networks and Transfer Learning. Tests message passing, graph pooling, expressive power (WL test), and transfer strategies (fine-tuning, LoRA, task arithmetic). Bridges structured-data reasoning with efficient model reuse.

Broad-coverage easy mock interview simulating a first-round screening. One question from each major DL cluster: foundations, training mechanics, optimization, regularization, CNN, RNN, attention, and modern paradigms. Build confidence before moving to harder mocks.

Second easy mock interview with fresh question selections across all 16 DL topics. Focuses on common definitional traps — same difficulty as Easy 01 but tests complementary concepts. Complete both easy mocks before attempting medium-level tests.

Applied reasoning mock interview at medium difficulty. 12 questions covering the full DL pipeline — from neural net fundamentals and training mechanics through modern architectures and transfer learning. Simulates the applied ML engineer interview loop at mid-level companies.

Second medium mock with fresh question selection. Emphasizes debugging intuition — gradient flow issues, normalization edge cases, optimizer quirks, and sequence modeling gotchas. Complements Medium Mock 01 by covering the remaining topic variants.

Senior-level hard mock interview covering edge cases, production failure modes, and nuanced architectural trade-offs. 15 questions spanning all major DL topics — expect questions on loss spikes, attention saturation, BatchNorm pitfalls, contrastive collapse, and LoRA forgetting. Target: senior ML engineer or research scientist roles.

Second hard mock with fresh question selection. Focuses on hardware-aware reasoning — FlashAttention, roofline analysis, quantization artifacts — alongside classic hard traps like exploding gradients, depthwise convolution latency, and knowledge distillation failure modes. Complements Hard Mock 01 with complementary concepts.

18-question elite screening for senior engineers and AI researchers. Every question targets production-grade judgment: Why does your training run spike at step 1200? Why does SimCLR collapse at small batch sizes? How does RoPE affect long-context extrapolation? Covers RMSNorm, DeepNorm, Neural ODE, MAE vs DINO, VAE posterior collapse, and weight tying. Requires deep architectural intuition, not just textbook recall.

18-question elite test targeting AI architect–level depth. Mix of hard MCQs and elevated medium questions that require synthesis: understanding why gradient checkpointing trades compute for memory, how LoRA rank affects task arithmetic interference, when depthwise separable convolutions hurt rather than help, and how link-prediction leakage poisons GNN benchmarks. Ideal as a final screening before staff/principal engineer or research scientist interviews.