Executive Summary
Twelve months, four stages, eight phases, fifty weeks. The plan takes an absolute beginner through programming fundamentals and mathematical bedrock (Months 1–2), into core computer-science mastery and classical ML built from raw NumPy (Months 3–5), through deep learning internals and production-grade C++ systems programming (Months 6–8), and finally into GPU kernel engineering, generative AI internals, distributed training, and staff-level production system design plus behavioral mastery (Months 9–12). Every phase pairs theory with a concrete, buildable artifact — nothing is passive reading.
| Stage | Months | Phase | Core Output |
|---|---|---|---|
STAGE 01 | 1–2 | Absolute Beginner Foundations | Python fluency, NumPy/Pandas, Linear Algebra, Calculus, Probability & Optimization from first principles |
STAGE 01 | 2 | Foundational Math | Hand-derived gradients, SVD/PCA, Bayesian reasoning, hand-rolled optimizers |
STAGE 02 | 3–4 | Data Structures & Algorithms | 150+ LeetCode Medium/Hard across arrays → DP, timed mock loops |
STAGE 02 | 5 | Classical ML From Scratch | Linear/Logistic Regression, Trees, RF, XGBoost mechanics, GMM/EM — vectorized NumPy only |
STAGE 03 | 6 | Deep Learning & Autograd | Hand-built autograd engine, PyTorch internals, CNN/ResNet, Transformer primitives |
STAGE 03 | 7–8 | C++ Systems & Concurrency | RAII, smart pointers, move semantics, multithreading, cache-aware performance engineering |
STAGE 04 | 9 | GPU Core & Triton | SM/warp architecture, tiled GEMM, softmax & reduction CUDA kernels, Triton |
STAGE 04 | 10 | Generative AI & LLMs | GQA/MQA/MLA, RoPE, MoE routing, LoRA/QLoRA, INT8/INT4 quantization, RLHF/DPO |
STAGE 04 | 11 | Distributed Systems & HPC | Ring-AllReduce, DDP/FSDP, Megatron TP, 1F1B pipelining, FlashAttention, classic system design |
STAGE 04 | 12 | Production ML Design & Leadership | Multi-stage rec-sys, HNSW vector search, vLLM/PagedAttention, RAG/Agents, MLOps, STAR stories |
The Daily Doctrine
Every single day — no exceptions — is split into exactly two execution blocks. This rhythm repeats for 350 consecutive days.
Core & Internals
45–60 MIN- Theory, mathematical derivations, and systems internals
- Read/derive by hand before ever touching code
- Whiteboard proofs, architecture diagrams, paper annotation
- Goal: understand why the mechanism works, not just that it works
Applied Execution
45–60 MIN- Hands-on Python / C++ / CUDA implementation
- LeetCode reps or architecture diagramming
- Ship something runnable and testable every day
- Goal: convert Block 1 theory into working, benchmarked artifacts
Build Pipeline — 4 Stages
Click a stage to jump to its weeks in the tracker below.
50-Week Execution Tracker
Check off each block as you complete it. Click the 🔗 icon to attach the specific articles, papers, videos, or repos you used that day. Progress saves automatically to your browser.
Data & Portability
Your state auto-saves locally. Export your JSON file to backup your progress or move it to a different device.
Export regularly (e.g., weekly) and keep the JSON file safe. Re-import it any time, on any device, to restore every checkbox and link.