Workbooks

AI/ML as a Force Multiplier

Why It Matters

Most “AI for labs” fails for one simple reason:

It tries to replace the human.

Real value comes from something else:

AI embedded at the implicit decision points where humans already work.

Calibrations.
Quality checks.
Drift detection.
Classification.
Go / no-go decisions.
Parameter nudges.

GaN Live

GaN Playback

Metrology Trifecta

Linear Composition Gradient

Workbooks turn those daily routines into deterministic, AI-assisted workflows — safely.

Not AI going wild.

AI doing the work you already do, but faster and more consistently.

Deterministic Agency

Workbooks are designed for bounded intelligence.

They run inside a structured environment with:

Explicit inputs
Defined outputs
Logged actions
Clear stop conditions
Human-in-the-loop control when needed

This is how you get real agency in industrial systems:

Determinism first.
Intelligence second.

A Built-In Framework + User Extensions

Workbooks ship with built-in templates and examples.

And they are extensible:

Versioned workbook definitions
User plugins supported
Standard patterns for common tasks
Reusable UI widgets and logic blocks

If you can define the workflow, you can systematize it.

UI + Logic

Each Workbook has two parts:

1) The UI Layer

Built from structured widgets:

Charts
Tables
Image viewers
Buttons / toggles
Status indicators

2) The Logic Layer

Implemented in code:

Reads internal state
Calls UnicornOne tools and commands
Triggers Recipes / Macros
Records results to immutable logs
Produces structured outputs

Workbooks are not “notes.”

They are executable workflows.

Calling External AI/ML Models

AI/ML models are called externally over TCP/IP.

But they behave like first-class tools:

Models load automatically when a workbook starts
Execution is bounded by workflow logic
Outputs are structured and logged
Decisions are explicit and traceable

This makes advanced AI usable without turning your control system into a science project.

Real-Time Examples

Frame-by-frame RHEED image analysis at ~14 ms latency.

Live classification
Automated annotation
Drift detection
Transition recognition

ROI intensity analysis + auto-annotation.

Extracts features continuously
Labels events
Creates clean datasets for training and improvement

The rule is simple:

If a human can do it — and historically had to —
you can train an AI to do it for you.

Safely.

The Philosophy

Workbooks are not “automation for the sake of automation.”

They are how you:

Reduce repetitive manual work
Increase consistency and throughput
Capture operational knowledge
Create clean datasets
Embed AI where it actually belongs

Less time calibrating.

More time imagining what’s possible.