Design Amazon Recommendation System
Candidate generation → ranking → re-ranking pipeline. Feature stores, A/B testing, real-time inference. Amazon signature SDI.
generalkafkaml-aisystem-design
ML / AI Engineer
Transformers, LLM internals, RAG, MLOps, model serving, evaluation, agent architectures. NVIDIA-flavored loops add CUDA depth.
Primary categories: ml-ai · system-design
Curriculum (11 topics)
Design Real-Time Autonomous-Vehicle Data Ingestion Pipeline
Vehicle → edge → cloud, sensor data compression, Kafka/Pulsar, hot/cold tiering, retraining loop.
cppkafkaml-aisystem-design
Design LLM Inference Serving (Triton + vLLM + PagedAttention)
Triton Inference Server, dynamic batching, KV-cache management, tensor parallelism, paged attention.
cudaml-aipythonsystem-design
CUDA Shared Memory & Memory Coalescing
GPU performance is bound by memory bandwidth. Shared memory + coalesced global access are the two highest-leverage CUDA optimizations. Required depth at NVIDIA and Tesla Autopilot.
cppcudageneral
Design Tesla Autopilot Data Pipeline (Auto-labeling + Retraining)
Auto-labeling, shadow mode, rare-event mining, training data curation, reprocessing.
cppcudaml-aisystem-design
Design Distributed Training for 10B+ Param Model on 128 H100/B200
Tensor + pipeline + data parallelism, NCCL all-reduce, FP8/BF16, ZeRO/FSDP, checkpointing. NVIDIA signature SDI.
cppcudaml-aisystem-design
Design Dynamic Home-Screen Recommendations
Per-profile row construction, ranking, freshness, real-time signal incorporation. 2025 Netflix-reported prompt.
generalml-aisystem-design
Design a GPU Cluster Scheduler (MIG + Gang)
Bin packing, MIG slicing, gang scheduling, K8s GPU operator, preemption, QoS. NVIDIA signature SDI.
cudakubernetessystem-design
Design Fault-Tolerant Distributed Training Pipeline
Checkpointing cadence, replica resurrection, NCCL recovery, elastic training. Survival in long-running jobs.
cudaml-aisystem-design
Design Netflix Recommendation Engine
Offline candidate generation + online ranking, "row-of-rows" homepage, A/B testing infra, personalization signals.
generalml-aisystem-design
STAR & STAR-L Method for Behavioral Stories
Structured story format used by every Mag7 behavioral round. Google extends to STAR-L (Learnings). Amazon expects 1 LP per story. Netflix tunes to culture pillars.
behavioralgeneral