Skip to content

ArdeaAI/playing-with-sand-in-black-boxes

Repository files navigation

Playing With Sand In Black Boxes

A multi-session workshop teaching AI/ML from first principles — from hand-built perceptrons to agentic systems.

Two tracks: a 3-session technical route and a 1-2 session non-technical user route.

Sessions

Session Topic Material Demos
01 Foundations Perceptrons, backprop from scratch, PyTorch intro session01-perceptron, session01-nn-scratch, session01-xor-pytorch
02 Modern Approaches MNIST CNN, custom models, fine-tuning, RAG, research frontiers session02-mnist, session02-custom
03 Agentic Systems ReAct agents, tool use, agentic SaaS landscape session03-agent

Setup

Requires Python 3.12 and UV.

# Install dependencies
uv sync

# Run the interactive menu
uv run sand

# Or run a specific demo directly
uv run sand session01-perceptron

CLI Commands

uv run sand                       Interactive menu
uv run sand session01-perceptron  Single-layer perceptron (AND, OR, XOR)
uv run sand session01-nn-scratch  Neural network from scratch (XOR)
uv run sand session01-xor-pytorch PyTorch XOR solution
uv run sand session02-mnist       CNN on MNIST digits
uv run sand session02-custom      Feedforward on Iris dataset
uv run sand session03-agent       Mock ReAct agent loop

Dependencies

  • rich — terminal UI (tables, panels, progress bars)
  • torch — PyTorch deep learning framework
  • torchvision — datasets and transforms (MNIST)
  • scikit-learn — datasets and preprocessing (Iris)

About

Brief history of ML / AI and a journey from home-baked perceptrons to agentic systems

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages