Curriculum

108 weeks, ten phases, one arc

Every instructional week has a module, a focus, a mode, and a core deliverable. There is no attendance-based completion - each week produces evidence.

108 weeks · 10 phases · 69 implement / 35 read+diagram / 4 deploy+benchmark · 13 recovery windows
Filter by phase
implement build & test working code read · diagram design, analyze & explain deploy · benchmark run on real infra & measure GPU weeks 75, 84, 87, 89
Week Phase Module Focus Mode Core deliverable
1 P1 Orientation and course OS Course setup, tracker, repo, weekly template implement Course dashboard + README
2 P1 Python engineering refresh Typing, testing, async basics implement CLI project skeleton with tests
3 P1 Numerical Python NumPy, vectorization, broadcasting implement Vector math utility library
4 P1 Reproducibility Git, notebooks, experiment tracking implement Reproducible notebook + Makefile
5 P1 Math diagnostic and induction Linear algebra, probability, hypothesis choice, and generalization implement Diagnostic report + gap map + paper lab
6 P1 Systems diagnostic HTTP, Docker, APIs, Linux basics implement Containerized hello-service
7 P2 Linear algebra I Vector spaces, span, basis, dimension implement Problem set + concept map
8 P2 Linear algebra II Linear independence, rank, nullity implement Solver exercises + notes
9 P2 Linear algebra III Inner products, norms, dual norms implement Norm/distance library
10 P2 Linear algebra IV Orthogonality, projections, Gram-Schmidt implement Projection demo
11 P2 Linear algebra V Linear transformations, change of basis implement Coordinate transform lab
12 P2 Linear algebra VI Eigenvalues, eigenvectors, diagonalization implement PCA-style mini lab
13 P2 Linear algebra VII PSD matrices, quadratic forms, Rayleigh quotient implement Proof + visualization
14 P2 Linear algebra VIII SVD, pseudoinverse, low-rank approximation implement Low-rank compression demo
15 P2 Numerical linear algebra I Matrix norms, condition numbers, perturbation implement Perturbation experiment
16 P2 Numerical linear algebra II LU, QR, Cholesky decomposition implement Decomposition library
17 P2 Numerical linear algebra III Iterative solvers, power iteration, Krylov methods implement Solver benchmark
18 P2 Numerical linear algebra IV Randomized SVD, tensor notation implement Randomized approximation lab
19 P3 Calculus I Partial derivatives, gradients, directional derivatives implement Gradient visualizer
20 P3 Calculus II Jacobians, Hessians, Taylor approximations implement Curvature lab
21 P3 Calculus III Chain rule on computational graphs implement Manual backprop worksheet
22 P3 Automatic differentiation Forward/reverse mode, JVPs, VJPs implement Tiny autodiff prototype
23 P3 Probability I Probability spaces, conditioning, Bayes theorem implement Probability problem set
24 P3 Probability II Random variables, distributions, transformations implement Simulation notebook
25 P3 Probability III Random vectors, covariance, correlation implement Covariance/correlation lab
26 P3 Probability IV LLN, CLT, delta method implement Monte Carlo convergence demo
27 P3 Statistics I MLE, MAP, estimator bias/variance implement Estimator comparison
28 P3 Statistics II Hypothesis testing, p-values, Type I/II errors implement Test implementation
29 P3 Statistics III Confidence intervals, bootstrap, permutation tests implement Bootstrap CI package
30 P3 Statistics IV Power, MDE, sample size, multiple testing implement Experiment-size calculator
31 P4 Retrieval math Cosine similarity, exact KNN implement Vector search mini-lab
32 P4 Top-K systems Memory-bounded KNN, streaming Top-K implement Heap-based Top-K service
33 P4 Ranking pipelines Dynamic Top-K, two-stage retrieval/reranking implement Reranking prototype
34 P4 Rate limiting I Sliding-window and token bucket rate limiters implement API limiter module
35 P4 Caching LRU cache and TTL cache implement Cache simulator with tests
36 P4 Text chunking Token-aware and recursive sentence chunking implement Chunking library
37 P4 PII handling PII span merging, regex/NER detection implement Redaction utility
38 P4 Calibration Expected Calibration Error, Brier score implement Reliability dashboard
39 P4 Ranking metrics NDCG, MRR, Recall@K implement Retrieval evaluator
40 P4 Batching Thread-safe inference batcher implement Concurrent batcher
41 P4 Resilience Retry, exponential backoff, jitter, circuit breaker implement Fault-tolerance kit
42 P4 Sketches Count-Min Sketch, Bloom filter implement Probabilistic data structures
43 P4 Cardinality/routing HyperLogLog, consistent hashing implement Distributed routing lab
44 P4 Load balancing Weighted round robin, rendezvous hashing implement Load balancer module
45 P4 IDs/traffic Snowflake IDs, leaky bucket, sliding-window counter implement Platform primitives lab
46 P5 System design framework Requirements, constraints, API design read · diagram Design doc template
47 P5 Data and APIs Data models, observability, contracts read · diagram API design exercise
48 P5 Rate limiting design Token bucket, sliding windows at scale read · diagram Rate limiter design writeup
49 P5 Scaling reads/writes Caching, replication, contention read · diagram Scaling pattern notes
50 P5 Large blobs Storage, CDN, media pipelines read · diagram Blob storage design
51 P5 Workflow systems Orchestration vs choreography read · diagram Workflow state design
52 P5 Schema evolution Backward compatibility, contracts read · diagram Schema migration plan
53 P5 Case studies I URL shortener, Dropbox read · diagram Two design docs
54 P5 Case studies II Ticketmaster, News Feed read · diagram Two design docs
55 P5 Case studies III WhatsApp, LeetCode read · diagram Two design docs
56 P5 Case studies IV Uber, web crawler read · diagram Two design docs
57 P5 Case studies V Ad click aggregator, payments read · diagram Trade-off analysis
58 P6 ML problem framing Objectives, constraints, iterative development read · diagram ML project charter
59 P6 Data quality Labeling, weak supervision, augmentation read · diagram Data quality checklist
60 P6 Leakage/contamination Detection, eval tracking, versioning read · diagram Leakage audit
61 P6 Offline evaluation Baselines, slices, confidence intervals implement Eval harness
62 P6 Online evaluation A/B tests, shadow deployment, canary, bandits read · diagram Rollout plan
63 P6 A/B calculator Lift, z-score, p-value, confidence intervals implement analyze_ab_test function
64 P6 Bootstrap methods Paired bootstrap for model comparisons implement bootstrap_ci package
65 P6 SRM detection Sample-ratio mismatch, chi-square diagnostics implement SRM detector
66 P6 Feature flags Rules, rollouts, sticky assignment implement Flag engine
67 P6 Experiment assignment Deterministic assignment, namespaces implement Assignment service
68 P6 Monitoring Drift, observability, four-layer monitoring read · diagram Monitoring dashboard spec
69 P6 Production failures Distribution shift, postmortems read · diagram Incident postmortem template
70 P7 LLM anatomy Transformer, tokenizer, matmul read · diagram Model anatomy notes
71 P7 LLM data pipelines Cleaning, deduplication, tokenization read · diagram Dataset card
72 P7 Fine-tuning SFT, LoRA, QLoRA, DoRA read · diagram Fine-tuning plan
73 P7 Alignment RLHF, DPO, ORPO, GRPO read · diagram Preference eval plan
74 P7 Distributed training DDP/FSDP, tensor/pipeline parallelism, ZeRO read · diagram Training architecture map
75 P7 Vector search HNSW, IVF-PQ, product quantization deploy · benchmark Index benchmark
76 P7 Retrieval patterns Hybrid search, query rewriting, chunking implement Retrieval lab
77 P7 Retrieval fusion BM25 + dense + reranker, RRF implement Fusion service
78 P7 Retrieval eval Query slices, macro metrics, regressions implement Slice evaluation report
79 P7 Prompt systems Prompt template renderer, version registry implement Prompt management tool
80 P7 Memory/context Conversation memory, context optimizer implement Memory service
81 P7 Agent foundations ReAct, plan-execute, tool calling, guardrails; agent harness (tool-approval gates, session persistence/resume, context compaction, sub-agent fan-out, sandboxing vs host-shell) implement Agent prototype
82 P8 GPU performance Compute/memory/overhead regimes, FlashAttention read · diagram Performance primer notes
83 P8 Inference arithmetic Roofline model, arithmetic intensity read · diagram Model intensity worksheet
84 P8 PagedAttention/vLLM KV paging, block tables deploy · benchmark vLLM deployment
85 P8 vLLM internals Scheduler, block manager, code paths read · diagram Internal anatomy notes
86 P8 vLLM metrics Running/waiting requests, latency histograms read · diagram Metrics schema
87 P8 Observability Prometheus/Grafana for serving deploy · benchmark Live inference dashboard
88 P8 Benchmarking Request-rate sweeps from 1 to 1000+ concurrent, saturation-point identification at each step read · diagram Benchmark methodology doc + one published optimization benchmark (vLLM or SGLang config; public gist or blog post)
89 P8 SGLang RadixAttention, prefix reuse deploy · benchmark SGLang comparison report + one published vLLM-vs-SGLang benchmark comparing the W84 and W89 pipeline configs
90 P8 Scheduling & MoE Execution Orca, continuous batching, chunked prefill, MoE all-to-all routing, Warp Decode, COMET overlap read · diagram MoE overlap schedule design doc + latency-hiding simulator
91 P8 Quantization FP8, AWQ, GPTQ, KV cache compression; INT4 (per-channel, per-group, block-wise) and BitNet-style 1.58-bit frontier read · diagram Quality/throughput table with INT4 row (QLoRA 4-bit normal-float as the production-ready path; BitNet as the research frontier)
92 P8 Long context & attention alternatives Medusa, EAGLE, StreamingLLM, KV eviction; linear attention (ELU+1), state-space (Mamba), delta-rule (DeltaNet/Gated DeltaNet), KDA/Kimi Linear; KV cache vs learned recurrence read · diagram Latency/memory lab
93 P8 Disaggregation DistServe, Splitwise, Mooncake, autoscaling; Kubernetes topology (KServe + KEDA + HPA) and edge deployment (ONNX, TensorRT-LLM, WebLLM) as alternative topologies read · diagram Disaggregated serving plan + depth-mode sub-deliverable: deploy the W84 pipeline to K8s with HPA-driven autoscaling on a synthetic spike AND port one vLLM config to ONNX/TensorRT-LLM for edge benchmarking (Modal A10 fallback if no edge hardware)
94 P9 Traffic routing Model router, kill switch, sticky rollout implement Router service
95 P9 Safe observability PII redaction, end-to-end request tracing implement Trace + safe logs
96 P9 Fan-out/streaming Partial failures, token stream multiplexing implement Resilient aggregator
97 P9 Quotas/priority Token budgets, gateway prioritization implement Quota/priority service
98 P9 Workflow execution DAG scheduler, idempotency implement Workflow engine
99 P9 Fairness/HA Per-tenant limits, leader election implement Tenant limiter + HA lab
100 P9 Async infrastructure Distributed work queue, config service implement Queue/config platform
101 P9 Model lifecycle Health monitor, registry/promotion implement Model ops pipeline
102 P9 Cost/safety/caching Semantic/embedding cache, injection detection, cost attribution implement Cost/safety platform
103 P10 Capstone planning Requirements, architecture, eval plan read · diagram Capstone design doc
104 P10 Capstone build I Core API, retrieval/agent flow implement Working prototype
105 P10 Capstone build II Evaluation, observability, guardrails implement Monitored prototype
106 P10 Capstone build III Load testing, optimization, cost tuning implement Performance report
107 P10 Mock interview week System design, coding, behavioral simulations read · diagram Mock interview scorecard
108 P10 Final review Portfolio, retrospective, job plan read · diagram Final portfolio + 30-day plan
Recovery windows

Deliberate catch-up, not extra topics

After these weeks the course pauses new content so you can consolidate, remediate gates, or rest. They are generated as part of the rhythm - a recovery week is never an additional topic.

After weeks: 6 · 14 · 22 · 30 · 38 · 46 · 54 · 62 · 70 · 78 · 86 · 94 · 102

Stackable programs

Three publishable checkpoints

W1–30Foundations

Math, numerical routines, and experimentation toolkit

Portfolio release · Week 30
W31–69Engineering + Systems

Tested primitives, system designs, MLOps tools, and postmortem

Portfolio release · Week 69
W70–108LLM Platform

LLM/RAG work, inference benchmark, platform layer, and capstone

Portfolio release · Week 108

100-module mathematics curriculum → · 30 engineering tasks + 40 practice drills →