spjaffolding 0.1.3

Produces rigorous advanced machine learning frameworks natively bypassing jinja

spjaffolding

A fast, strictly-typed Python CLI application that scaffolds advanced, highly modular Machine Learning frameworks. Designed for Principal Software Engineers and Research teams who demand robust architecture without the pollution of traditional templating engines (like Jinja).

By leveraging Python's Typer, rich, questionary and dynamic programmatic string injection coupled with the Rust-backed speed of uv, spjaffolding ensures deterministic environment resolution and clean, readable codebases.

Key Features

• Jinja-Free Static Generation: Avoids external templating engines by utilizing localized textwrap.dedent blocks, keeping the scaffolding tool lintable (ruff) and type-checked (mypy).
• Interactive Multi-Select CLI: Uses a beautiful macOS-native-looking, rich-styled checkbox interface (via questionary) to let users dynamically toggle specifically the modules they need.
• Dynamic Feature Architecture: Additive folder structures and pyproject.toml dependencies based on your selections:
  • gpu, viz, data_prep, model_training, scraping, vector_db, llm_orchestration.
• Reusable Presets: Easily load preconfigured bundles like --preset numerical_ml or --preset llm_scraping. Save your own custom selections to ~/.config/spjaffolding/presets.json for future runs, and view them anytime using --list-presets.
• Mode-Driven CLI Controller: Generates a boilerplate script integrating an argparse/Typer Finite State Machine (FSM), natively supporting sequential chained execution (e.g., -m prepare_train_evaluate).
• uv Environment Orchestration: Subprocess execution of uv init and uv sync locally bypasses pip dependency hell with uncompromised lockfile reproducibility. Injecting strict dependency clusters like gpu, viz, and test based on PEP 621/735.

Installation

You can install spjaffolding globally to bootstrap ML projects anywhere on your filesystem using uv tool:

uv tool install . --force

Quick Start

To view all available predefined and custom presets, formatted beautifully in your terminal:

spjaffold --list-presets

Invoke the scaffolding wizard to use the interactive checkbox menu:

spjaffold --tool-name "phaze" --description "High Energy Physics anomaly detector"

Or run rapidly via a built-in or custom saved preset:

spjaffold --tool-name "phaze" --description "HEP anomaly detector" --preset "numerical_ml"

This generates the complete structural footprint based on your toggled features (e.g., numerical_ml preset):

phaze/
├── README.md
├── pyproject.toml
├── src/
│   └── phaze/
│       ├── __init__.py
│       ├── cli.py       <-- Automatically bound to the selected domain FSM!
│       ├── data_prep/
│       ├── models/
│       └── training/
└── workspaces/

Within the scaffolded framework, execute your modes sequentially:

uv run phaze -m "prepare_train" -p default_workspace default_project

Developer Guide

spjaffolding itself maintains high structural integrity. Check its code standard via:

uv sync --all-extras
uv run ruff check src
uv run mypy src
uv run pytest tests