Learn AI — XploreAI Interactive Learning Hub

🎯

Supervised Learning

Teach by example — labeled data trains models to predict.

Supervised Learning

You train a model by showing it thousands of examples where you already know the correct answer — like showing a child thousands of pictures labeled "cat" or "dog" until they can identify new ones. The model makes predictions, checks if it was right, and adjusts itself. After millions of adjustments, it becomes accurate enough to deploy.

Real-World Story — Gmail Spam Filter

Google trained its spam filter on hundreds of millions of emails humans had already labeled "spam" or "not spam." The model learned patterns — certain words, sender domains, link structures. Now it filters 99.9% of spam without any human checking each email. That's supervised learning at scale.

How It Works — Step by Step

1

Collect labeled training data (emails marked spam/not spam)

2

Feed it to the model — it makes a prediction on each example

3

Compare the prediction to the correct label — calculate the error

4

Adjust the model's internal weights to reduce that error

5

Repeat millions of times until accurate enough to deploy

Key Terms

Training Data

Labeled examples used to teach the model

Label

The correct answer for each training example

Overfitting

Model memorizes training data but fails on new data

Accuracy

% of correct predictions on unseen data

⚡

XploreAI uses this: Claude AI classifies patient messages — understanding whether someone wants to book, cancel, or ask a question — based on supervised learning on millions of labeled conversation examples.

🔍

Unsupervised Learning

No labels needed — find hidden structure in raw data.

Unsupervised Learning

No labeled data, no correct answers — the model gets raw data and finds hidden patterns on its own. It's like being dropped into a foreign city with no map and figuring out the layout yourself. Most of the world's data is unlabeled, so this is incredibly powerful.

Real-World Story — Spotify Discover Weekly

Spotify never asks you to label songs by genre. It uses unsupervised clustering on 600 million users' listening patterns, finding that people who listen to Artist A also listen to Artist B — without anyone defining the relationship. It groups users into "taste clusters" and recommends what similar listeners love. The result feels eerily personal.

How It Works — Step by Step

1

Feed raw, unlabeled data to the algorithm (e.g., user listening history)

2

Algorithm calculates similarity between all data points

3

Groups similar points into clusters (K-means, DBSCAN, etc.)

4

Each cluster represents a discovered pattern (a taste group, a customer type)

5

New data points are assigned to the nearest cluster automatically

Key Terms

Clustering

Grouping similar data without predefined categories

Centroid

The calculated center point of a cluster

Dimensionality Reduction

Compressing data to fewer variables (e.g., PCA)

Anomaly Detection

Finding data points that don't fit any cluster

⚡

XploreAI uses this: Analysing clinic patient behaviour — peak booking times, common queries, drop-off points — unsupervised clustering finds segments without anyone manually defining them.

🧠

Neural Networks & Deep Learning

Interconnected layers of nodes learn like a brain.

Neural Networks & Deep Learning

A neural network is connected mathematical nodes, loosely inspired by brain neurons. Each node takes inputs, applies a transformation, and passes the result forward. Stack many layers and you get "deep learning" — each layer automatically learns a different level of abstraction: edges → shapes → features → objects. No human programs these rules; the network discovers them from data.

Real-World Story — iPhone Face ID

Face ID uses a deep neural network. Setup: 30,000 infrared dots build a 3D face map. Unlock: a new scan is compared by the network. It accounts for glasses, beards, hats, even aging — because it learned the 3D geometry of your unique face, not a 2D photo. Works in the dark. Can't be fooled by a photograph.

How It Works — Step by Step

1

Input data (image pixels, audio, text) enters the first layer

2

Each node multiplies the input by a "weight" and adds a "bias"

3

An activation function (like ReLU) decides if that node "fires"

4

Result passes to the next layer, which learns higher-level features

5

Errors flow backward (backpropagation) to adjust all weights

Key Terms

Weight

A number controlling how much a connection matters

Backpropagation

Adjusting weights based on prediction errors

ReLU

Activation function: outputs 0 if negative, else passes value through

Layer

A group of nodes that transform data together

⚡

XploreAI uses this: Claude claude-opus-4-6 — the brain of the WhatsApp bot — is a deep neural network with hundreds of billions of parameters. Every intelligent response is a forward pass through this massive network.

💬

Large Language Models (LLMs)

Predict the next token — at trillion-parameter scale.

Large Language Models (LLMs)

LLMs are neural networks trained on enormous amounts of text to predict the next word in a sequence. "Large" = hundreds of billions of adjustable numbers. Repeat the prediction billions of times across all of human writing, and something remarkable emerges: language, logic, facts, and reasoning — not programmed, but learned.

Real-World Story — Claude Writes a Booking Confirmation

When a patient sends "Rahul Desai" to confirm their slot, Claude has processed the entire conversation — all messages, the clinic's system prompt, the confirmed time. It generates "✅ Appointment Confirmed! Patient: Rahul Desai, Date: Monday 24 Mar, Time: 7:20 PM…" token by token — not from a template, but predicting the most contextually appropriate next word each time.

How It Works — Step by Step

1

Text is split into "tokens" (roughly ¾ of a word each)

2

Each token is converted to a high-dimensional vector (its embedding)

3

The Transformer processes all tokens, learning which relate to which (attention)

4

Model predicts a probability distribution over all possible next tokens

5

Highest-probability token is selected and appended; process repeats

Key Terms

Token

Basic unit of text an LLM processes (~¾ of a word)

Context Window

How many tokens the model can "see" at once

Temperature

Controls creativity: high = varied, low = focused

Prompt Engineering

The art of writing inputs that get the best outputs

⚡

XploreAI uses this: Claude claude-opus-4-6 — Anthropic's most capable model — is the intelligence layer. Every patient conversation is processed by this LLM in real-time, generating contextually perfect responses for any query.

👁️

Computer Vision

CNNs decode pixels into meaning, patch by patch.

Computer Vision

Computers see images as grids of numbers (pixel values 0–255). Computer vision models learn to find meaningful patterns in those numbers — edges, shapes, textures, objects — trained on millions of labeled images. Deep learning transformed this field completely: today's models detect cancer in X-rays better than radiologists.

Real-World Story — AI X-Ray Reading at Manipal Hospitals

A deep learning model trained on 1.2 million chest X-rays flags pneumonia, tuberculosis, and lung nodules in under 3 seconds — and highlights exactly which region raised the alert. Radiologists review the AI's finding rather than scanning from scratch. In a 500 X-ray/day hospital, this cuts reporting time by 60% and catches cases tired eyes miss.

How It Works — Step by Step

1

Image fed as a grid of pixel values (e.g., 224×224×3 for RGB)

2

Convolutional layers scan in small patches, detecting local features

3

Pooling layers shrink the representation while keeping key patterns

4

Multiple conv+pool layers stack up, each learning more abstract features

5

Final layers output a class label or bounding box coordinates

Key Terms

CNN

Convolutional Neural Network — standard architecture for images

Feature Map

Output of a conv layer — compressed detected features

Object Detection

Locating and classifying multiple objects in one image

Transfer Learning

Fine-tuning a pre-trained model for a new task

⚡

Coming to XploreAI: Prescription reading — a patient photographs their prescription and the AI automatically reads the doctor's name, medication, and dosage to pre-fill booking forms.

🎮

Reinforcement Learning

Learn by trial, error, and reward — like training a dog.

Reinforcement Learning

RL is how AI learns through experience — like a child learning to ride a bike by trying, falling, adjusting, and eventually succeeding. No labeled dataset. An "agent" takes actions, receives rewards or penalties, and gradually learns which actions lead to the best outcomes. Given enough exploration, RL agents often discover strategies no human would think to try.

Real-World Story — How ChatGPT Became Helpful (RLHF)

After pre-training on text, GPT-4 was fine-tuned using Reinforcement Learning from Human Feedback. Human trainers rated pairs of AI responses — "which is more helpful?" Those ratings became the reward signal. The RL algorithm then optimised the model to score highly with humans. This is how a raw text predictor becomes a helpful, harmless assistant. Claude uses the same technique.

How It Works — Step by Step

1

Agent observes the current state of the environment

2

Picks an action based on its current policy (strategy)

3

Environment transitions to a new state and gives a reward signal

4

Agent updates its policy to favour actions that gave higher rewards

5

Over thousands of episodes, policy converges to near-optimal behaviour

Key Terms

Agent

The AI making decisions (bot, robot, language model)

Environment

The world the agent acts in (game, conversation, robot space)

Reward Function

The signal telling the agent how well it's doing

Policy

The agent's strategy: given state X, take action Y

⚡

XploreAI roadmap: Future bot versions will use RL principles to automatically learn which response styles lead to successful bookings vs. drop-offs — continuously improving conversion rate without manual tuning.

Learn Artificial Intelligence

Core AI & ML Concepts

Supervised Learning

Unsupervised Learning

Neural Networks & Deep Learning

Large Language Models (LLMs)

Computer Vision

Reinforcement Learning

Latest AI Releases

Learning Path