Week 5: Backtracking (finish), Graphs
Oct 13–19 · Target times: Easy ≤ 12m · Medium ≤ 25m · Hard ≤ 45m
Daily routine: 75 min DSA (timed) + 10 min syntax typing + 35 min second block. Talk out loud. No AI autocomplete. Over cap → 48h redo queue.
DSA problems (75 min, timed)
Second block (35 min) — System Design + CS: ML System Design (your zone)
Model serving (batching, replicas, autoscaling), feature store, monitoring/drift, A/B testing. Most important: serving at scale + monitoring. Practice: design a model-serving / recommendation pipeline (ByteByteGo & Hello Interview ML episodes). Outcome: inference API for 10k QPS with a latency budget.
DSA problems (75 min, timed)
Second block (35 min) — ML/DL: Serving & optimization concepts
Dynamic batching, quantization (INT8/FP16), ONNX/TensorRT, GPU utilization/contention, latency vs throughput. Most important: your GPU-contention/Triton scheduling story, formalized.
DSA problems (75 min, timed)
Second block (35 min) — System Design + CS: ML System Design (your zone)
Model serving (batching, replicas, autoscaling), feature store, monitoring/drift, A/B testing. Most important: serving at scale + monitoring. Practice: design a model-serving / recommendation pipeline (ByteByteGo & Hello Interview ML episodes). Outcome: inference API for 10k QPS with a latency budget.
DSA problems (75 min, timed)
Second block (35 min) — ML/DL: Serving & optimization concepts
Dynamic batching, quantization (INT8/FP16), ONNX/TensorRT, GPU utilization/contention, latency vs throughput. Most important: your GPU-contention/Triton scheduling story, formalized.
DSA problems (75 min, timed)
Second block (35 min) — System Design + CS: ML System Design (your zone)
Model serving (batching, replicas, autoscaling), feature store, monitoring/drift, A/B testing. Most important: serving at scale + monitoring. Practice: design a model-serving / recommendation pipeline (ByteByteGo & Hello Interview ML episodes). Outcome: inference API for 10k QPS with a latency budget.
🚀 PROJECT 2 — Inference server + benchmark (KICKOFF weekend)
Repo; serve 2–3 models (FastAPI multi-model). Build benchmark harness (Locust/custom) for latency + throughput under concurrency. Milestone: baseline server + benchmark running.
Also: 1 system-design practice out loud. Redo any DSA problem that went over cap (48h queue).