Learn — Deep Guides¶
These are the primary reading materials. Each guide takes a single lecture and explains it in plain English — the problem it solves, an intuition (usually with an analogy), a worked example, the formula with meanings, common misunderstandings, and how it shows up in the exam.
Every guide ends with 5 practice MCQs and a quick-reference summary.
How to read one
Budget 20–35 minutes per guide. Read straight through, then attempt the 5 questions at the end WITH THE ANSWERS COVERED. Whatever you got wrong or guessed on — reread just that concept block before moving to the next lecture.
Recommended reading order¶
The lectures build on each other. If you have time, read them in order. If you're time-crunched, start with Lecture 11 (vector indexing) — it alone accounts for 6 of the 24 sample paper questions.
| # | Guide | What you'll learn | Approx. read time |
|---|---|---|---|
| 1 | ML foundations | Precision, recall, F1, confusion matrix, ML pipeline stages | 25 min |
| 2 | Neural networks & word embeddings | MLP → CNN → RNN → LSTM → Word2Vec → architecture families | 30 min |
| 3 | Transformers & attention | Self-attention, Q/K/V, positional encoding, multi-head | 30 min |
| 4 | LLM decoding & APIs | Greedy/top-k/top-p/temperature, OpenAI Chat Completions, HF Inference vs Ollama | 25 min |
| 5 | Structured outputs & evaluation | JSON schema, LogitsProcessor, BLEU / ROUGE / METEOR / BERTScore | 30 min |
| 6 | AI safety & CIA triad | Confidentiality/Integrity/Availability, prompt injection, jailbreaks | 25 min |
| 7 | Advanced prompting | Zero-shot / Few-shot / CoT / Self-Consistency / ReAct / Tree-of-Thoughts | 30 min |
| 8 | Prompt security & APE | Levenshtein distance, log-probability scoring, robustness | 25 min |
| 9 | RAG fundamentals | Retrieve → augment → generate, naive vs advanced vs agentic | 25 min |
| 10 | Dense retrieval & rerankers | Bi-encoder vs cross-encoder, ColBERT, two-stage retrieval | 25 min |
| 11 | Vector indexing & PQ math | Product Quantization, IVF, HNSW, query routing, RBAC gateway | 35 min — highest yield |
| 12 | Agentic RAG | Observe-Think-Act loop, tool use, iterative retrieval | 25 min |
What "learn" means here¶
Each guide follows the same pattern per concept:
- What problem this solves — why anyone invented this
- Intuition — usually an analogy from everyday life
- Worked example — a small numeric or textual example you can follow
- Formula with meaning — every symbol explained
- Common misunderstandings — the traps students fall into
- Exam angle — how this exact concept appears in questions
If a concept doesn't click on first pass, reread just the intuition and the worked example — they're the fastest path to a mental model.