This repository is a collection of declarative configuration files used to manage my Python development environments across multiple devices. Adopting the "Infrastructure as Code" philosophy, I ensure high consistency across physical machines, remote servers, and containers.

Guiding Principle: Rebuild over Patch

When dependencies change, I follow the "Destroy -> Rebuild" principle. Manual incremental modifications via CLI are discouraged. All changes must be implemented within configuration files to guarantee idempotency and environment purity.

Important Update: This repository is currently undergoing an architectural migration. Following a deep evaluation, I have decided to shift the focus of environment management from traditional Conda/Pip to a modern, declarative toolchain centered around uv.

• Extreme Performance: Having understood zero-copy linking and global caching mechanisms, I believe uv will become the de facto management standard for the Python community.
• Deterministic Reproduction: Conda and Pip struggle to achieve perfect reproduction via config files alone. The uv ecosystem finally solves idempotent builds through modern lockfile mechanisms.
• Scientific Computing Compatibility: To maintain compatibility with research requirements, I use pixi for Conda-protocol support. Other tools remain under observation.

This repository employs Declarative Environment Management: whenever a configuration changes, the environment is destroyed and rebuilt from scratch.

I maintain a series of base configurations tailored for different workflows:

• agent_env: Agent development environment centered on LangChain and LangGraph.
• dl_env: Deep Learning environment, focused on PyTorch model design and training.
• ds_env: Data Science research, including analysis and visualization.
• llm_env: Dedicated environment for LLM execution, deployment, and training.
• spider_env: Web crawling and automation tools.
• web_env: Base environment for backend development.

For project-specific configurations, please refer to the conda_specific_projects/ directory.

# Sync project dependencies using uv
uv sync

# Initialize environment using pixi
pixi shell -e dl_env

The following are examples based on conda. While Conda has a long-standing influence in scientific computing, I plan to migrate these to uv and pixi in the future.

First, list the existing virtual environments.

conda env list

If an old environment exists, remove it.

conda deactivate
conda remove --name ${env_name} --all

Rebuild the environment using the configuration file.

conda env create -f ./environment.yaml

Finally, verify that the dependencies are correctly installed.

conda activate ${env_name}
conda list

These commands are also documented in the conda_environments/conda_run.sh file.

• Introduce uv to replace native pip.
• Introduce pixi for declarative management of research environments.
• Migrate all configurations to pyproject.toml and pixi.toml.
• Integrate Remote Development based on Docker Dev Containers for a "total solution."

Other toolkits I maintain that you might find useful:

• Python-Environment-Configurations : Environment configurations for Python toolchains (Pip, Conda, UV, etc.).
• Deployment-Toolkit : A collection of raw OS scripts (Shell/PowerShell).
• CLI-Wrapper : A safer, more user-friendly Python wrapper for complex CLI tools.