AI / ML — 8 Modules · Beginner to Advanced

Learn Artificial Intelligence
& Machine Learning.

From the fundamentals of AI to cutting-edge Generative AI — this track covers everything you need: ML algorithms, deep learning, NLP, computer vision, reinforcement learning, and a complete DS & AI cheat sheet.

Start with Machine Learning AI/ML Interview →
⬡8 topics covered ⬡Beginner to advanced ⬡Interview-focused ⬡Includes cheat sheet
8 Modules AI → Cheat Sheet
25+ Interview Qs In the AI/ML track
6 Core Fields ML · DL · NLP · CV · RL · GenAI
1 Cheat Sheet DS & AI reference
Foundations

What is Artificial Intelligence?

Coined by John McCarthy in 1956. AI is the design of intelligence in an artificial device — machines that acquire, understand and apply knowledge to achieve goals, able to learn and solve problems on their own. It borders mathematics, computer science, philosophy, psychology, biology and cognitive science.

Artificial Intelligence any technique that mimics intelligent behaviour — search, logic, planning, agents Machine Learning learns patterns from data instead of hand-coded rules Deep Learning multi-layer neural networks Generative AI LLMs · diffusion
Think · Human-like

Cognitive Science approach

Model human cognition precisely enough to simulate it — the goal is reproducing the reasoning steps a human follows, not just the answer.

Think · Rationally

Laws of Thought approach

Formal logic and inference mechanisms that are provably correct and guarantee an optimal solution when one exists.

Act · Human-like

Turing Test approach

A machine is intelligent if a human interrogator can't tell its answers from a person's — behaviour over mechanism.

Act · Rationally

Rational Agent approach

Build agents that do the best possible action given what they know — the dominant view in modern AI (Russell & Norvig).

Foundations

Intelligent agents

An agent perceives its environment through sensors and acts on it through actuators: agent = architecture + program. A task environment is specified as PEAS — Performance measure, Environment, Actuators, Sensors.

Type 1

Simple reflex

Acts only on the current percept via condition→action rules. Ignores history; needs a fully observable environment.

Type 2

Model-based reflex

Keeps an internal state + a model of "how the world works" — handles partially observable environments.

Type 3

Goal-based

Has an agenda: searches and plans toward an explicit goal, choosing actions by whether they get closer to it.

Type 4

Utility-based

When many paths reach the goal, picks the best one by maximising a utility ("happiness") function.

Environment properties that shape agent design:

Fully vs partially observable Deterministic vs stochastic Episodic vs sequential Static vs dynamic Discrete vs continuous
Foundations

Classical AI: search & game playing

Before ML, AI solved problems by searching a state space — start state, goal state, and operators between them. Still the backbone of pathfinding, planning and game AI.

Uninformed (blind) search

BFS · DFS · Iterative Deepening

No knowledge of how far the goal is. BFS explores level-by-level (optimal for unit costs, memory-hungry); DFS dives deep (light on memory, can loop); iterative-deepening DFS combines both — DFS memory with BFS completeness.

Informed (heuristic) search

Hill Climbing · Best-First · A*

Uses a heuristic h(n) estimating distance to goal. Hill climbing greedily improves and gets stuck on local maxima/plateaus; A* orders by f(n) = g(n) + h(n) and is optimal when h never overestimates (admissible).

Adversarial search

Minimax + Alpha-Beta pruning

Two-player games: MAX picks moves maximising the score, MIN minimises it, evaluated bottom-up from leaf positions. Alpha-beta pruning skips branches that can't change the outcome — same answer, far fewer nodes.

Constraint satisfaction

CSP: Backtracking + Local Search

Variables, domains, constraints (map colouring, Sudoku, scheduling). Backtracking assigns values and undoes dead ends; local search repairs a complete assignment step-by-step.

Foundations

Knowledge representation & reasoning

How an agent stores facts and derives new ones.

Logic

Propositional & First-Order Logic

Propositional logic: whole statements joined by ¬ ∧ ∨ → ↔. First-order logic adds objects, relations and quantifiers (∀, ∃) — expressive enough to state "every student passed some exam".

Inference

Forward vs Backward chaining

Forward chaining: start from known facts, fire rules until the goal appears — data-driven. Backward chaining: start from the goal and work back to supporting facts — goal-driven (Prolog-style).

Probabilistic reasoning

Bayes' theorem

P(H|E) = P(E|H) · P(H) / P(E)

Update belief in hypothesis H after seeing evidence E. Foundation of Naive Bayes classifiers, spam filters and medical diagnosis — reasoning under uncertainty instead of hard true/false logic.

Curriculum

All 8 AI / ML modules

Follow the path top-to-bottom or jump to any topic directly.

Module 01
Machine Learning
Supervised Unsupervised Algorithms Model Evaluation
Core ML algorithms — linear regression, decision trees, SVMs, clustering, and how to train, evaluate, and tune models effectively.
Intro level →
Module 02
Deep Learning
Neural Networks CNNs RNNs Backpropagation
Neural networks from perceptrons to deep architectures — feedforward, convolutional, and recurrent networks with activation functions and optimizers.
Mid level →
Module 03
NLP & LLMs
Tokenization Transformers BERT LLMs
Natural language processing — from text preprocessing to transformers, attention mechanisms, BERT, GPT, and how modern LLMs work.
Mid level →
Module 04
Computer Vision
Image Processing Object Detection YOLO Segmentation
Teaching machines to see — image classification, object detection with YOLO, image segmentation, and real-world CV pipelines.
Mid level →
Module 05
Reinforcement Learning
Agents & Rewards Q-Learning Policy Gradient
Learning through interaction — agents, environments, rewards, Q-learning, and policy-based methods that power game AI and robotics.
Advanced →
Module 06
Generative AI
GANs Diffusion Models Prompt Eng. RAG
The frontier — GANs, VAEs, diffusion models, prompt engineering, RAG pipelines, and how tools like ChatGPT and Stable Diffusion are built.
Advanced →
Module 07
DS & AI Cheat Sheet
Quick Reference Formulas Interview Ready
One-page reference for data science and AI — key formulas, algorithm comparisons, complexity cheat sheet, and interview quick-prep.
Reference →
Live Algorithm Visualizations
Step through algorithms interactively — no installs, runs entirely in your browser
6 Algorithms
K-Means Clustering Unsupervised
K-Means iteratively assigns points to the nearest centroid (★), then recomputes centroids as cluster means. Converges when assignments stop changing. Sensitive to initial placement — hit New Data for different starting positions. Uses k = 3 clusters.
Linear Regression Supervised
Ordinary Least Squares finds the line y = mx + b minimising the sum of squared residuals (shown as red vertical lines). The best-fit line minimises Σ(yᵢ − ŷᵢ)². Click the canvas to add new points, then re-fit.
Neural Network Forward Pass Deep Learning
Forward Pass — activations flow left → right. Each node computes a weighted sum then applies a non-linearity (ReLU in hidden layers, Sigmoid on output). Edge thickness encodes weight magnitude. Bright nodes = high activation. Architecture: [3 → 5 → 4 → 2].
Decision Tree Supervised
Decision Trees recursively split data by the feature giving the best Gini impurity or Information Gain. Each node is a yes/no question; leaves are predictions. Green = Accept, Yellow = Review, Red = Reject.
PCA — Principal Component Analysis Dimensionality Reduction
PCA finds orthogonal directions of maximum variance. PC1 captures the most variance; PC2 is orthogonal to it. Points can be projected onto fewer dimensions — compressing data while retaining structure. Hit New Data for a fresh dataset.
Gradient Descent Optimization
LR 0.04
Gradient Descent updates parameters by moving in the direction of steepest descent: θ ← θ − α · ∇L. Each dot is one step. Adjust LR — too high causes overshooting; too low causes painfully slow convergence. The pink dot is the current position on the loss surface.
Check yourself

Quick Quiz

Five questions on the map of the field: agents, classical search, and where learning fits.

1. PEAS specifies a task environment. What does it stand for?
2. A simple reflex agent acts only on the current percept. What does that require of its environment?
3. Iterative-deepening DFS is used because it combines which two properties?
4. What does first-order logic add over propositional logic?
5. How do AI, machine learning and deep learning relate?
← Home