v5.0.4 · PowerShell 5.1 · Educational AI Framework · Learn by doing

VBAF is a hands-on educational framework for learning artificial intelligence and reinforcement learning concepts — written entirely in PowerShell 5.1 that ships with every Windows PC.

No Python. No Jupyter. No cloud dependencies. Just open PowerShell and start learning.

What you can learn with VBAF:

• How neural networks learn through backpropagation
• How Q-learning agents discover optimal strategies without being told the rules
• How Deep Q-Networks (DQN) scale reinforcement learning to complex problems
• How multiple agents compete and cooperate in shared environments
• How to normalise, scale and clean data before feeding it to an AI model

Why PowerShell?

Because the code is readable. Every function in VBAF is written to be understood — not just executed. You can open any .ps1 file and see exactly what the algorithm is doing, line by line. That is the point.

# Install
Install-Module VBAF -Scope CurrentUser

# Navigate to your working folder
cd "C:\Users\\OneDrive\WindowsPowerShell"

# Load everything
. .\VBAF.LoadAll.ps1

# Run the XOR example -- the classic neural network benchmark
cd examples\01-XOR-Network
. .\Run-Example-01.ps1

# Watch a Q-learning agent learn castle sequences
cd ..\02-Castle-Learning
. .\Run-Example-02.ps1

# See competing market agents emerge pricing strategies
cd ..\03-Market-Simulation
. .\Run-Example-03.ps1

Start here and work through in order:

Each example folder contains a Run-Example-XX.ps1 launcher -- just run that.

Three interactive tools for guided learning:

# Console teacher -- 6 topics, press Enter to advance
Start-VBAFTeach

# Jump to one topic directly
Start-VBAFTeach -Topic "DQN"
Start-VBAFTeach -Topic "QLearning"
Start-VBAFTeach -Topic "Enterprise"

# Interactive experiment station -- pick algorithm, configure, watch it train
Start-VBAFPlayground

# Jump to one playground directly
Start-VBAFPlayground -Algorithm "DQN"
Start-VBAFPlayground -Algorithm "Enterprise"
Start-VBAFPlayground -Algorithm "Supervised"

XOR is the classic test of whether a neural network can learn non-linear patterns. A linear model cannot solve it -- you need at least one hidden layer.

. .\VBAF.LoadAll.ps1
cd examples\01-XOR-Network
. .\Run-Example-01.ps1

# Expected output:
# XOR Truth Table:
#   0 XOR 0 = 0  (predicted: 0.02)
#   0 XOR 1 = 1  (predicted: 0.97)
#   1 XOR 0 = 1  (predicted: 0.96)
#   1 XOR 1 = 0  (predicted: 0.03)
# Epochs: 847  Loss: 0.008

If you see predictions close to 0 and 1 -- the network learned. That is backpropagation working.

A Q-learning agent generates castle sequences and discovers that variety is rewarded. No strategy is programmed. The agent finds it through trial, error and reward signals.

cd examples\02-Castle-Learning
. .\Run-Example-02.ps1

# Watch the Q-table grow:
# Episode   1 | Q-Table:  10 entries | Exploit:  0.0%
# Episode  50 | Q-Table: 286 entries | Exploit: 11.6%
# Episode 100 | Q-Table: 383 entries | Exploit: 22.2%

This is reinforcement learning in its purest form -- learning from reward signals, not from labelled examples.

Once you understand the foundation phases (1-9), VBAF includes 14 enterprise pillars built on the same core -- each one a working DQN agent solving a real IT automation problem.

The learning ladder: Phase 1-9 -- Foundation: understand HOW agents learn Phase 10-27 -- Enterprise: see WHAT agents can do

• Windows 10 or 11
• PowerShell 5.1 (included with Windows -- no install needed)
• No Python, no Jupyter, no cloud account, no dependencies

VBAF is designed to be taught. The docs/teaching/ folder contains:

• A full 4-week course outline with session plans and lab exercises
• Exam questions at beginner, intermediate and advanced levels
• Semester plan and teaching notes

# The interactive teacher -- works in any classroom
Start-VBAFTeach

# Students experiment independently
Start-VBAFPlayground

MIT License -- see LICENSE for details. Free to use, teach, modify and share.

Henning · Roskilde, Denmark 🇩🇰 Built with Claude (Anthropic) · PowerShell ISE · PS 5.1

"The best way to understand AI is to build it yourself -- line by line."