Hamilton.cr is a lightweight Crystal framework for building Telegram bots, designed to simplify the development of interactive and high-performance bots. It supports handling commands, text messages, and rich media payloads (stickers, videos, GIFs, etc.), while offering built-in features like context storage for stateful conversations, asynchronous processing for scalability, and signal-based task control for managing long-running operations. The framework minimizes boilerplate and is easily extensible with custom handlers, making it ideal for both simple and complex bot applications.

A FastAPI-inspired framework for building REST APIs in Crystal. Fossil.cr simplifies API development by leveraging Crystal’s macro system to generate boilerplate code for request/response handling. It provides a clean HTTP::Server wrapper with intuitive endpoint routing, supporting path/query/form/body parameters, dependency injection, and custom content types. The framework minimizes manual request/response operations, allowing developers to focus on business logic.

tgsh is a Telegram bot template that provides secure shell access to your server via Telegram. Designed for developers and system administrators, it allows you to execute short and long-running commands, interact with processes using POSIX signals, and manage files directly from your Telegram chat. The bot supports Python script execution, real-time logging, and custom log file management, making it ideal for remote server administration. Note: Avoid using it on cloud VMs, as it may violate Terms of Service.

UBPE is a Universal Byte-Pair Encoding (BPE) tokenizer designed for general sequences, including quantized signals and text. Unlike traditional BPE tokenizers, which focus primarily on text, UBPE supports custom alphabets and large token spaces, making it ideal for compressing 1D signals, time-series data, or multilingual text. The library provides an optimized classic BPE implementation for general sequences, as well as a novel approach that generates multiple encoding variants for a single sequence, allowing users to select the best representation based on TF-IDF scores. UBPE also includes Python-native and Cython (C++20) backends for high performance, making it suitable for both research and production use.

A PyTorch-based library for integrating wavelet transforms into neural networks. The library enables seamless integration of wavelet-based feature extraction and processing within deep learning pipelines, supporting both 1D and 2D wavelet transforms for applications like signal processing, image analysis, and time-series data augmentation. By leveraging PyTorch’s autograd system, WaveletNN-PyTorch allows for end-to-end training of wavelet-augmented neural networks, combining the strengths of wavelet analysis with modern deep learning techniques.

Machine Learning Library from scratch. Provides data normalization methods, many optimization methods, linear models, decision trees, and clusterization algorithms.

Laurent polynomials with basic arithmetic, evaluation and substitution. Library is useful for wavelet analysis and was developed during a university course on the subject.

Little Julia projects and guides in Pluto.jl notebooks.

Polynomial ring with basic arithmetic, parsing, formatting, prime tests and generating.

Tool for synchronization multilingual documentation in your project.

Regular Expressions library in pure C. RegExes are compiled into finite state machines and can be reused for multiple queries.

Library that proposes a new P2P network topology that enables users to connect multiple devices to one peer. The topology provides an opportunity to implement distributed applications over it.