Python Packages

I created and maintain the following Python packages:

echoes

A scikit-learn compatible package for Machine Learning with Echo State Networks

  • 🤖 Machine learning with Echo State Networks
  • 🧱 High level API, scikit-learn compatible API, eg fit-predict interface
  • 💨 Fast! (Thanks to numba ✨)
  • 🐍 pip install echoes

kuramoto

Classical model to study synchronization phenomena

  • 🤹 Implementation of the Kuramoto model on graphs.
  • 🐍 pip install kuramoto

ser

Susceptible-Excited-Refractory, a dynamical model of spreading excitations

  • 🚥 Implementation of classic SER model on graphs
  • 🏃 Powered by numba for speed.
  • 🐍 pip install ser

Python Course

A friendly, hands-on introduction to Python.
This is a free, short introduction to the Python programming language, emphasizing practical over theoretical aspects and with a focus on data-related tasks.

cronex.nvim

A Neovim Plugin to render in-line, human-readable cron expressions. Cronex is kind of a “client” that allows the Neovim user to integrate and customize “servers” that actually parse the expressions (“explainers”) in a flexible fashion.

pypkg.guru

Python Package Guru is a search engine for Python Packages that offers a minimalistic and fast exploration of the official Package Index (PyPI).

crates.guru

Semantic Search of Rust Crates. Explore the official crates.io site but using Natural Language.

kubeflow-pipeline-uv

A simple implementation of a custom component that helps to manage dependencies of Kubeflow components with uv. Check the blog post here or watch my presentation at PyData Berlin 2025 here:

TorchServe

TorchServe is a popular framework to deploy PyTorch-based neural networks. I have fixed bugs in Dockerfile and build scripts. I have also implemented CI (GitHub) workflows for continuous testing of docker images and containers.

torchserve-docker

Container images with specific TorchServe and Python version working out-of-the-box. This is a GitHub-Action-based cron job that watches for TorchServe releases and automatically generates and pushed the images to Dockerhub.

logger-bot

logger-bot is a lean Telegram workout assistant that turns voice notes into database training logs.
It’s a small composed system: chat interface in front, ML extraction in the middle, and a queryable data store behind it. Here’s an architecture diagram:

flowchart LR
    user([User])
    telegram(Telegram Bot)
    server([Server])
    extractor([Speech2Text])
    logger([Logger])
    db([Database])

    user -- voice --> telegram
    telegram -- polling --> server
    server -- audio --> extractor
    extractor -- text --> server
    server -- text --> logger
    logger --entry--> db

On top of logging, it exposes lightweight command handlers for health checks, last-set lookup, and read-only SQL queries over your workout history. For example you can run a query with a command like this:

Other Contributions

Minor tests/documentation/bug-fixes

For more details, check out my GitHub Profile.