Meet AI & Orientation
Understand what AI is, how the field evolved, and the mental model for the journey.
Meet AI: The Journey to Modern Intelligence
The human story, historical milestones, AI winters, and the convergence that led to modern LLMs
Explore the driver-and-maps analogy, the history of AI from its foundations through Transformers and LLMs, why the current era accelerated, and what language models can—and cannot—do.
Foundations
Build the mathematical and intuitive foundations needed to understand models and learning.
Essential Math & Intuitive Foundations
The 4 applied mathematical pillars that power vectors, attention, softmax, and gradient learning
Vectors & GEMM matrix multiply, calculus loss gradients, probability softmax, and high-dimensional cosine similarity
Language & Model Architecture
Follow language through tokens, vectors, embeddings, recurrent networks, attention and Transformers.
Understand Language
How AI turns raw text into numbers it can reason about
The atomic unit of all language models
Why text must become numbers and what those numbers mean
Dense vectors that capture meaning — the foundation of modern AI
Understand Context
How models learned to remember and focus on what matters
The first attempt at sequential memory — and why it struggled
Gated memory cells that solved the vanishing gradient problem
Dynamic focus — the breakthrough that made Transformers possible
Process Context Efficiently
Self-Attention, QKV, and the Transformer — the engine of modern AI
Every token attending to every other token simultaneously
Query, Key, Value — the mathematical heart of attention
Parallel attention heads capturing different relationship types
How Transformers know word order without recurrence
The complete architecture that runs every modern LLM
Model Training
Understand how data, model design, pretraining and post-training produce a trained model.
03.2 · Explore all 14 model-training stages →Build Language Models
Pretraining, LLMs, generation — how language models actually work
Next-token prediction — the one task that enables everything
How models learn from billions of tokens
What makes a language model large — and powerful
How models produce text — sampling, temperature, decoding
Adapting a pretrained model to specific tasks and behaviors
Shaping model behavior with human preference signals
Model Inference
Follow a prompt through a deployed model to generated output using fixed trained weights.
04.1 · Open the interactive 11-stage inference pipeline →AI Systems Engineering
Connect models to knowledge, context, prompts, tools, agents and external systems.
Give Models Knowledge
Embeddings, vector databases, and RAG — making models knowledgeable
Text → vectors for semantic search and retrieval
Storing and searching millions of embeddings at speed
Retrieval-Augmented Generation — giving models access to your data
Give Models Capabilities
Prompting, tools, agents, and MCP — making models act
The craft of instructing language models effectively
Function calling — giving models access to external actions
LLMs that observe, plan, act, and loop autonomously
Model Context Protocol — the standard for connecting agents to tools
Mastering what goes into the context window — the most important LLM skill
Production & Optimization
Evaluate, secure, deploy and optimize AI applications and model-serving systems.
Build, Evaluate & Ship AI Applications
The applied engineering layer between understanding model capabilities and operating dependable AI products
Understand where models are published, hosted, run locally and accessed through compatible interfaces.
Choose a model based on task fit, quality, latency, cost, context, deployment constraints and portability.
Connect models to software reliably using APIs, SDKs, streaming, structured responses and predictable error handling.
Use coding assistants and agents as part of a verifiable engineering workflow rather than trusting generated code blindly.
Extend the RAG foundations into a dependable retrieval pipeline with measurable grounding and traceable sources.
Measure quality and operational behavior with repeatable tests, traces, cost signals and regression checks.
Design AI services that remain reliable under load, recover from provider failures and stay within cost budgets.
Protect applications that combine model outputs, retrieved documents, user data and tools that can take actions.
Make Models Efficient
SLMs, quantization, distillation, and inference optimization
Running powerful AI locally and on edge devices
Reducing model size by compressing weight precision
Training small student models to mimic large teacher models
KV cache, batching, and serving optimizations
Advanced AI & Frontier
Explore advanced alignment, reasoning, multimodal models and emerging research directions.
Advanced Alignment
RLHF, reasoning models, chain-of-thought, and multi-modal — making models safer and smarter
Teaching models to think step-by-step before answering
Models that see, hear, and understand beyond text
Emerging AI
Diffusion, vision transformers, mixture of experts, and constitutional AI — the frontier
Iterative denoising to generate images, audio, and more
Applying the Transformer architecture directly to image patches
Scaling model capacity by routing tokens to specialized experts
Principle-based self-critique and revision to align model behavior
High-Impact AI & Frontier Areas
Key developments, curated links, YouTube videos, and breakthrough papers driving the bleeding edge of AI
A curated playground of cutting-edge research, YouTube deep dives, and expert articles
Career & Portfolio — Supporting Path
Apply the curriculum through projects, open source contributions and technical growth. This supports all eight tracks rather than adding a ninth track.
AI Career Growth & Portfolio
Turn knowledge into demonstrable engineering skill, public work, product outcomes and long-term growth
Turn learning into visible evidence through useful projects, reproducible demos, documentation and contributions.
Connect technical depth to problem discovery, communication, interviews, product thinking and leadership.
Interactive Labs — Practice Across All Tracks
Hands-on labs reinforce the concepts in the curriculum. They are practice experiences, not separate numbered learning tracks.
Interactive Live Labs
Step-by-step simulators and animated playgrounds — run every concept in your browser, connect to a local Ollama model for live inference
19 stages — every transformer concept from raw text to final answer, one step at a time
Animated stage-by-stage walkthrough — QKV vectors, scaled dot-product, softmax matrix, multi-head heads, all wired live
8-step deep dive into attention math — token inputs, QKV projection matrices, dot products, scaled softmax, and full multi-head assembly
Interactive sentence flow — choose tokens, observe animated query-to-key signal routes, inspect QKV vectors, and run signal through a 6-layer architecture stack
8 stages of transformer block computation — Embeddings, Multi-Head Attention, Residual connections, LayerNorm, and Feed-Forward MLP
Next-token prediction pipeline — temperature, logits, softmax, and sampling all wired up live
Autoregressive token-by-token generation with temperature scaling, Top-k, Top-p nucleus filtering, and probability sampling
Type any text — watch it split into pieces, count characters, whitespace tokens, and UTF-16 units in real time
Edit numeric vectors and compute dot product, Euclidean distance, and cosine similarity — exact results every time
Context changes representation — compare how "bank" differs in river vs finance context with exact cosine calculations
Think → Act → Observe → Repeat — deterministic mock environment to inspect tool calls, errors, and empty results