I implemented Deep Q-Networks in pure PowerShell 5.1 — and connected them to real Windows enterprise data

Yes, really. No Python. No TensorFlow. No cloud.

VBAF is a full reinforcement learning framework written in PowerShell 5.1 classes. It includes DQN, PPO, A3C, Q-Learning, CNNs, RNNs, AutoML, MLOps — all from scratch.

The latest release (v3.0.0) adds Enterprise Automation agents that read real Windows data:

# Agent watching live CPU and learning to optimize
Invoke-VBAFResourceOptimizerTraining -Episodes 100

# Agent reading Event Logs and learning alert routing
Invoke-VBAFAlertRouterTraining -Episodes 100

# Agent learning job scheduling from Task Scheduler patterns
Invoke-VBAFJobSchedulerTraining -Episodes 100

Results: +292% improvement on scheduling, +230% on alert routing vs random baselines.

The PS 5.1 constraints made this genuinely hard (no operator overloading on typed arrays, no closures, single-threaded class methods) — but that made solving it more interesting.

Install-Module VBAF
GitHub: https://github.com/JupyterPS/VBAF

Happy to write a deep-dive on the DQN implementation in PS 5.1 if there’s interest!