Start Here
Start here Is This Book Beginner-Friendly? A Difficulty Map Short answer: yes for Days 2–8, and yes with patience for Days 9–15. Every day opens with a plain-words Big Picture, uses analogies, and works small numbers by
Start here
Is This Book Beginner-Friendly? A Difficulty Map
Short answer: yes for Days 2–8, and yes with patience for Days 9–15. Every day opens with a plain-words Big Picture, uses analogies, and works small numbers by hand before showing code. The steepest climbs are Days 9–10 (backpropagation), Day 13 (attention) and the GenAI bridges, because each stacks several new ideas at once. Use this map to slow down at the right moment instead of feeling lost.
| Day | Level | Be comfortable with first | Pace | If you feel lost |
|---|---|---|---|---|
| 1 · Prerequisites | ●○○○ ●○○ Easy Easy | Basic Python syntax and school math | 1 day | Redo the worked problems, then the practice questions |
| 2 · NumPy | ●○○○ ●○○ Easy Easy | Python lists and loops (Day 1) | 1 day | Re-read the “axis” trick and shapes, p. 58 |
| 3 · Pandas | ●○○○ ●○○ Easy Easy | NumPy basics (Day 2) | 1 day | Print df.head() and df.info() after every step |
| 4 · Regression + GD | ●●○○ ●●○ ○ | Averages; slope of a line | 1 day | Line-by-line gradient descent, p. 105 |
| 5 · Overfitting + CV | ●●○○ ●●○ ○ | Day 4 (loss, fitting) | 1 day | Redraw the train / validation / test split from memory |
| 6 · Metrics + Trees | ●●○ ●●○○ ○ | Day 5 | 1–2 days | Fill a confusion matrix by hand for 10 fake predictions |
| 7 · k-NN, PCA, Embeddings | ●●○○ ●●○ ○ | Vectors and distance (Days 1–2) | 1–2 days | Compute cosine similarity of two 2-D vectors by hand |
| 8 · Telco Project | ●●○○ ●●○ ○ | Pandas (Day 3) and metrics (Day 6) | 1–2 days | Run each step on 5 rows first, then on the full data |
| 9 · NN From Scratch ★ | ●●● ●●●○ ○ | Matrix multiply; “slope” idea of a derivative | 1–2 days | Print .shape of every array in the forward pass |
| Day | Level | Be comfortable with first | Pace | If you feel lost |
|---|---|---|---|---|
| 10 · Backpropa gation ★ | ●●●○ ○ | Day 9 | 1–2 days | Follow the worked example, p. 193 |
| 11 · PyTorch | ●●●○ ○ | Days 9–10 | 1–2 days | Print .shape after every line |
| 12 · MNIST + CNN | ●●●○ ○ | Day 11 | 1–2 days | Track shapes through the CNN table, p. 250 |
| 13 · Attention ★★★ | ●●●● ● | Days 9–12 | 2–3 days | Two-vector example, p. 262 |
| 14 · Capstone | ●●○○ ○ | Everything above | 1–2 days | Pick the smallest project option |
| 15 · GenAI Bridges | ●●●○ ○ | Day 13 (attention) | 1–2 days | Run the toy programs before reading the theory |
Three rules that make the difference 1. Never skip the Big Picture at the start of a day. It is the map; the details are the streets. 2. Type the code, do not paste it. Typing forces you to notice what each line does. 3. Explain it aloud in plain words. If you cannot, go back one step in the dependency map (p. 11) rather than pushing on.
Honest limits
“15 days” is an intensive path, not a promise of mastery; plan extra days for Days 6 and 9. Run each PyTorch example once and compare your output with the text.
How to Use This Book
Use the detailed day-by-day notes as the main technical reference. Each day now begins with a short Big Picture overview and ends with a Quick Recap and a checkpoint. Whenever a concept feels too fast, unfamiliar, or disconnected from GenAI, slow down and re-read the Big Picture first.
The learning loop
(1) Understand → (2) See an analogy → (3) Work a tiny example → (4) Code it → (5) Break it → (6) Fix it → (7) Explain it → (8) Check yourself → (9) Move on
What “15 days” really means This is an intensive 15-day curriculum, not a promise of mastery in 15 days. A beginner may need extra revision days — especially for backpropagation, PyTorch, CNNs, attention, and Transformers.
How each day is laid out
- Big Picture — the idea in plain words (read this first).
- Detailed notes — explanations, worked examples, code, and assignments.
- Quick Recap — a compact page you can revise from.
- Checkpoint — five questions to test yourself before moving on.
Symbols you will see
- ✓ = correct / recommended, ✗ = wrong / avoid.
- ★ marks topics that matter most for GenAI.
- Blue boxes = key ideas · amber boxes = things that commonly go wrong.
The dependency map
Each topic builds on the one before it. If something feels confusing, the problem is usually a missing prerequisite from earlier in this chain.
- Python basics
- NumPy + Pandas
- ML concepts + data workflow
- Loss + gradient descent
- Evaluation + generalization
- Neural networks + backprop
- PyTorch
- CNNs
- Embeddings
- Tokenization
- Attention
- Transformers
- LLM fundamentals
- RAG
- Agents
- Tool calling
- MCP
- Evaluation
- Security
- FastAPI
- Docker
- LLMOps
When to slow down If you cannot explain a concept in simple words, stop and review its prerequisites instead of memorizing the formula.
How content is labeled
Core = must understand before moving on (every lesson, worked problem and “Move on when you can…” list). Practice = skills you build by coding (sections marked ★ ASSIGNMENT, your-turn problems, and the Assignment & Project Guide). Optional advanced = read after the core feels solid (Day 15, GenAI Bridges, and the supplements in the Appendix).
New to machine learning, or lost on a hard day? Read the Beginner’s Companion at the back of the book. It retells the hardest ideas (neural networks, backpropagation, PyTorch, CNNs, attention, RAG and agents) with everyday pictures and tiny numbers, and adds a gentle study plan and a self-check quiz. Every hard day also has a small “Stuck?” link in its page footer.
Open the Companion (p. 460) →