Getting Started

This project is under active development. Installation steps and CLI commands may change.

Prerequisites

• Python 3.11 or later
• A reMarkable tablet (for real stroke data; synthetic data works without one)

Installation

rm-gregg uses a modular dependency system. Install only what you need:

# Core only (extraction + preprocessing)
pip install -e .

# With ML support (classification, training)
pip install -e ".[ml]"

# With API server
pip install -e ".[api]"

# With feedback engine
pip install -e ".[feedback]"

# Everything (ML, API, feedback, visualization, PDF, dev tools)
pip install -e ".[all]"

Basic Usage

Extract stroke data from a .rm file

rm-gregg extract notebook.rm -o strokes.json

This parses the reMarkable binary format and outputs structured stroke data as JSON, including per-point coordinates, pressure, tilt, speed, and direction.

Generate synthetic training data

rm-gregg synthetic --unit 1 --count 100 -o data/synthetic/

Generates parameterized Gregg shorthand primitives (circles, lines, curves) with realistic variation for training without a tablet.

Train a stroke classifier

rm-gregg train data/synthetic/ --unit 1 --model rf -o models/unit1.pkl

Trains a Random Forest or SVM classifier on extracted features from stroke data.

Start the API server

rm-gregg serve --port 8000

Launches a FastAPI server for stroke classification and feedback.

Project Structure

src/rm_greg/
├── models.py          # Core data models (Stroke, Point, etc.)
├── cli.py             # Command-line interface
├── ingest/            # Extract stroke data from .rm files
├── preprocessing/     # Normalize, segment, extract features
├── training/          # Stroke classifiers and datasets
├── feedback/          # DTW comparison, scoring, feedback generation
├── api/               # FastAPI server
└── synthetic/         # Synthetic training data generation

What’s Next

See the Architecture page for details on how the pipeline works, or the Roadmap for planned development.