⚙️ Applied AI · Part 2 · Hands-on with Code
Applied AI
The hands-on sequel: go from understanding AI to building it in real PyTorch. Train a GPT from scratch, fine-tune with LoRA, and wire up RAG with a vector database. Every module pairs real, copyable code with interactive explainers.
▶ Start with Module 1
New here? Do the AI Fundamentals course first — it teaches the concepts this one builds in code.
HALF 1
Build & Train Models
The real version of the fundamentals simulator — build a model from raw tensors all the way up to a working GPT, then learn how to adapt existing models efficiently.
MODULE 1
The Environment: PyTorch & Tensors
Colab vs local, install, tensors, shapes, tensor operations, CPU/GPU.
pytorchtensors
→
MODULE 2
Autograd: The Engine of Learning
Gradients, the computation graph, .backward() — backprop in real code.
gradientsbackprop
→
MODULE 3
The Training Loop
The 5-step loop, loss functions, SGD/Adam, learning rate, overfitting, dropout & eval mode.
trainingoptimizers
→
MODULE 4
Datasets & Tokenizers
Hugging Face datasets, tokenization (BPE), batching, DataLoader, embeddings.
datatokenizers
→
MODULE 5 · ⭐
makemore — Build a Name Generator
Your first real model: bigram → neural net → generates new names, character by character.
build itsampling
→
MODULE 6 · ⭐⭐
nanoGPT — Build GPT From Scratch
The summit: tokens, embeddings, self-attention, transformer blocks — a real GPT in code.
attentiontransformer
→
MODULE 7
Fine-tuning a Pretrained Model
Specialize GPT-2 on your data: small LR, freezing layers, checkpoints — the practical way.
fine-tunecheckpoints
→
MODULE 8
LoRA, PEFT & Quantization
Fine-tune huge models cheaply: LoRA adapters, quantization, QLoRA on one GPU.
LoRAQLoRA
→
↓
HALF 2
Build Systems With Models
Shift from building models to building with them — the ML foundations of search, retrieval, and memory that power real AI apps.
🎓 By the end, you'll be able to build AND deploy AI — in real code:
✓ PyTorch & tensors
✓ Autograd / backprop
✓ The training loop
✓ Datasets & tokenizers
✓ makemore
✓ nanoGPT from scratch
✓ Fine-tuning
✓ LoRA / QLoRA
✓ Embeddings search
✓ Vector databases
✓ RAG
✓ Memory systems
Hybrid & hands-on: every module shows real, copyable code next to interactive explainers. The animations are illustrations — to actually run the code, copy it into Google Colab or your own machine. You're never asked to just watch; you build.
How to use this map: Half 1 builds bottom-up (each module needs the last), so go in order. Half 2's modules are more standalone. Click any card to jump in, or hit "Start with Module 1" and use the Continue → button at the bottom of each page.