Projects

Overview: Kalliope is a framework designed to help developers create their own personal assistants by providing a platform to attach signals (vocal orders, scheduled events, MQTT messages, etc.) to predefined actions called neurons. This allows users to easily build their own bots without needing to write code. Key features: - Easy-to-use framework for personal assistant development - Supports multiple languages and Linux distributions - Compatible with Raspberry Pi and other Linux-based systems - No need for coding; users can choose and compose existing neurons or write their own - Runs on various Linux distributions, including Raspberry Pi - Available as a web app and Android app Tech stack: Python Why it matters: Kalliope simplifies the process of creating a personal assistant by providing a user-friendly interface and a modular system of neurons that can be easily integrated into a project. This makes it accessible to developers of all skill levels, allowing them to quickly prototype and iterate on their assistant ideas.

Overview: Anyquery is a powerful SQL query engine that allows developers to run SQL queries on a wide variety of data sources, including files, databases, and apps. It is built on SQLite and uses plugins to extend its functionality, making it a versatile tool for integrating with various LLMs and MySQL-compatible clients. Key features: - Supports querying files, databases, and apps. - Built on SQLite and uses plugins to extend functionality. - Connects to LLMs to enable them to access data. - Acts as a MySQL server, allowing SQL queries from MySQL-compatible clients. Tech stack: Go Why it matters: Anyquery simplifies the process of integrating data from multiple sources into a single SQL query, making it easier for developers to work with complex data structures and access them through a unified interface. This tool is particularly useful for developers working with a variety of applications and databases, as it provides a single, consistent way to interact with their data.

### Overview: 百聆是一个基于语音识别、大语言模型和语音合成技术的开源语音对话助手，旨在提供高质量的语音对话体验。通过集成DeepSeek R1等优秀大模型和接入openClaw，百聆能够实现低延迟、无需GPU的高效运行，适用于各种边缘设备和低资源环境。 ### Key features: - **流畅对话体验**: 低延迟、不卡顿，几乎像真人对话一样自然。 - **轻量级部署**: 无需高端硬件，甚至不需要GPU，通过优化，可本地部署。 - **模块化设计**: ASR、VAD、LLM和TTS模块相互独立，可以根据需求进行替换和升级。 - **智能记忆功能**: 具备持续学习能力，能够记忆用户的偏好与历史对话，提供个性化的互动体验。 - **工具调用能力**: 灵活集成外部工具，用户可通过语音直接请求信息或执行操作，提升助手的实用性。 - **任务管理**: 高效管理用户任务，能够跟踪进度、设置提醒，并提供动态更新，确保用户不错过任何重要事项。 ### Tech stack: Python ### Why it matters: 百聆不仅是一个“能说话”的助手，更是一个“能做事”的助手。通过OpenClaw作为核心工具调用引擎，百聆能够处理复杂任务、外部工具编排和高阶Agent能力，提升语音助手的实用性。OpenClaw使得百聆从“聊天机器人”升级为“行动型助手”，具备强大的任务管理和工具集成

Overview: Quiver is an interactive visualization tool for understanding convolutional neural network (CNN) architectures. It simplifies the process of analyzing CNN models by providing a user-friendly dashboard that can be launched with a single line of code. Key features: - Launches a visualization dashboard with a single line of code - Supports Keras models and allows exploration of layer activations on input images - Provides options for customizing output classes, top predictions, and input image storage - Supports custom data mean and standard deviation for improved visualization Tech stack: JavaScript Why it matters: Quiver simplifies the process of understanding and analyzing CNN models, making it easier for developers to interpret and debug their models.

### Overview: Extractous is a fast and efficient solution for extracting content and metadata from various documents types such as PDF, Word, HTML, and many other formats. It is written in Rust and is designed to be highly performant and memory-efficient, leveraging Rust's strengths in performance and memory management. ### Key Features: - **Fast Processing:** Extractous is 25x faster than the popular unstructured-io library. - **Native Execution:** It achieves this speed by using native code execution and leveraging Rust's multi-threading capabilities. - **Cross-Platform Compatibility:** Extractous supports a wide range of file formats and can be easily integrated into projects using bindings for multiple languages, including Python. ### Tech Stack: - **Rust:** The core of Extractous is written in Rust, a language known for its performance, memory safety, and efficient use of CPU resources. - **Apache Tika:** For file formats not natively supported by the Rust core, Extractous compiles the well-known Apache Tika into native shared libraries using GraalVM ahead-of-time compilation technology. These shared libraries are then linked to and called from the Rust core, ensuring pure native execution. ### Why It Matters: Extractous was born out of the need for a faster, more efficient solution for content extraction from unstructured data. It offers a comprehensive solution in Rust with bindings for multiple languages, providing a significant improvement in performance and resource utilization compared to existing libraries like unstructured-io.

Overview: This repository contains a collection of advanced context engineering techniques and patterns designed to enhance the quality and predictability of Claude Code agents. Key features: - Simple to use: Easy installation without dependencies. - Token-efficient: Prefers command-oriented skills to minimize unnecessary context. - Quality-focused: Each plugin focuses on improving specific areas. - Granular: Only installs necessary plugins. - Scientifically proven: Based on validated techniques and patterns. - Open-standards: Skills adhere to the agentskills.io specification. Tech stack: TypeScript Why it matters: This toolkit provides developers with high-quality context engineering techniques that can significantly improve the performance and reliability of Claude Code agents. Repository: NeoLabHQ/context-engineering-kit GitHub stars: 1783 GitHub description: A collection of advanced context engineering techniques and patterns for Claude Code, OpenCode, Cursor, Antigravity, and other agents. Includes CodeRabbit as an open-source alternative. # Context Engineering Kit A curated set of advanced context engineering techniques and patterns aimed at enhancing the quality and predictability of Claude Code agents. Key Features: - **Simple to Use**: Easy to install and use without any dependencies. Contains automatically used skills and self-explanatory commands. - **Token-Efficient**: Prefers command-oriented skills to minimize unnecessary context. - **Quality-Focused**: Each plugin focuses on improving specific areas. - **Granular**: Only installs necessary plugins. - **Scientifically Proven**: Based on validated techniques and patterns. - **Open-standards**: Skills adhere to the agentskills.io specification

Overview: This repository is part of the Deep Learning course co-developed by Yandex, Higher School of Economics, and Skoltech. It provides supplementary materials for the course, including lecture notes, seminar discussions, and homework assignments. Key features: - Lecture notes on deep learning topics - Seminar discussions on neural networks and optimization methods - Homework assignments for practice and assessment Tech stack: Jupyter Notebook Why it matters: This course is designed to supplement the Deep Learning course taught at Yandex, Higher School of Economics, and Skoltech, providing additional resources and support for students. Repository: yandexdataschool/Practical_DL Primary language: Jupyter Notebook GitHub stars: 1780 GitHub description: DL course co-developed by YSDA, HSE and Skoltech GitHub issues: 1780 Telegram chat room: [chat room](https://t.me/+xPR3yh35AeViY2My) (russian-speaking HSE track) Grading rules: [this page](https://github.com/yandexdataschool/Practical_DL/wiki/Homeworks-and-grading-(HSE)) Installation guide: [installing pytorch](https://github.com/yandexdataschool/Practical_DL/issues/6) Contributors: Victor Lempitsky, Victor Yurchenko, Vadim Lebedev, Dmitry Ulyanov, Fedor Ratnikov, Oleg Vasilev, Arseniy Ashukha, Mikhail Khalman, many students and volunteers

### Overview: This project is a PyTorch implementation of Faster R-CNN, focusing on real-time object detection. It includes forward pass, RoI pooling layer, training on VOC2007, and TensorBoard support. ### Key Features: - **Forward Pass**: Real-time detection capability. - **RoI Pooling Layer**: Implemented on both CPU and GPU. - **Training**: VOC2007 dataset training. - **TensorBoard Support**: Visualizes training progress. ### Tech Stack: PyTorch ### Why it Matters: This project serves as a practical guide for re-implementing Faster R-CNN and understanding real-time object detection with PyTorch. It includes a detailed implementation and training process, making it useful for developers looking to build their own object detection models.

### Overview: NLP-Models-Tensorflow gathers machine learning and Tensorflow deep learning models for NLP problems, providing simplified Jupyter Notebook implementations. ### Key features: - Abstractive Summarization - Chatbot - Dependency Parser - Entity Tagging - Extractive Summarization - Generator - Language Detection - Neural Machine Translation - OCR - POS Tagging - Question-Answers - Sentence pairs - Speech-to-Text - Spelling correction - Squad Question-Answers - Stemming - Text Augmentation - Text Classification - Text Similarity - Text-to-Speech - Topic Generator - Topic Modeling - Unsupervised Extractive Summarization - Vectorizer - Old-to-Young Vocoder - Visualization - Attention ### Tech stack: Jupyter Notebook ### Why it matters: This repository simplifies complex NLP models and provides beginner-friendly implementations, making it ideal for research and educational purposes.

### Overview: MEX is a team memory solution designed for engineers and their AI agents. It captures the architecture, decisions, requirements, and handoffs of projects in a shared repository, allowing for easy review, collaboration, and knowledge sharing. ### Key Features: - **Inbox and Relay Skills**: Setup includes the official Inbox and Relay skills for Claude Code or Codex integration. - **Standalone Skill Installation**: Alternative for existing MEX projects. - **Repository Integration**: Keeps team's architecture, decisions, requirements, and handoffs alongside the code. - **Git as Sharing Layer**: Engineers and their agents can build on shared context, review proposed changes, and carry work between sessions and teammates. ### Tech Stack: TypeScript ### Why It Matters: MEX provides a durable home for knowledge, ensuring that what one engineer learns becomes context for the next teammate, enhancing collaboration and project efficiency.

Overview: OpenCode plugin that provides persistent memory for AI coding agents using local vector database technology. Key features: - Local Turso/libSQL database with native vector search - Persistent project memories - Automatic user profile learning - Unified memory-prompt timeline - Full-featured web UI - Intelligent prompt-based memory extraction - Multi-provider AI support (OpenAI, Anthropic) - 12+ local embedding models - Smart deduplication - Built-in privacy protection Tech stack: TypeScript Why it matters: This plugin enables long-term context retention for AI coding agents, improving productivity and accuracy in development tasks.

### Overview: WeMM-Embedding is a family of universal multimodal embedding models developed by the WeChat Vision team at Tencent. It provides unified representations for text, images, videos, visual documents, and interleaved multimodal inputs, achieving state-of-the-art performance across multiple benchmarks covering diverse tasks and domains. ### Key features: - **Unified Representations**: WeMM-Embedding generates a single unified representation for multimodal inputs, enabling cross-modal understanding and retrieval. - **State-of-the-Art Performance**: Achieves leading performance on various benchmarks, including image, video, visual document, and text tasks. - **Versatile Inputs**: Supports text, images, videos, visual documents, and interleaved multimodal inputs. - **Hugging Face Compatibility**: Models are compatible with Hugging Face's Transformers library, making them easily accessible and reproducible. ### Tech stack: - **Python**: The primary programming language used for developing and deploying WeMM-Embedding models. ### Why it matters: WeMM-Embedding is crucial for applications requiring multimodal understanding and retrieval, such as image captioning, video summarization, and visual document analysis. Its unified representation capabilities make it an essential tool in the field of computer vision and natural language processing. ### Repository: Repository: **Tencent/WeMM-Embedding** GitHub Stars: **1770** GitHub Description: **WeMM-Embedding is a family of universal multimodal embedding models by the WeChat Vision Team at Tencent, supporting multimodal understanding and retrieval.**

### Overview: This repository is a curated list of the top 200 deep learning GitHub repositories, sorted by the number of stars. The query used with the Github search API includes keywords like "deep-learning", "CNN", "RNN", and "convolutional neural network". ### Key Features: - **TensorFlow**: An open-source machine learning framework for everyone. - **Keras**: A high-level neural networks API that works on top of TensorFlow. - **OpenCV**: A powerful library for computer vision tasks. - **DeepLearning-500-questions**: A book that provides answers to common questions in deep learning, including probability, linear algebra, machine learning, and computer vision. - **TensorFlow-Examples**: A collection of TensorFlow tutorials and examples suitable for beginners. ### Tech Stack: Python ### Why it Matters: These repositories are essential for anyone interested in deep learning, providing comprehensive resources and practical examples to enhance their understanding and skills in the field.

### Overview: Relaticle is an open-source CRM system designed for developer-led teams, AI-forward startups, and SMBs. It integrates with AI agents such as Claude, ChatGPT, or open-source models over the MCP protocol, providing a native AI agent experience. The system is built with Laravel 13, Filament 5, and PHP 8.5, offering a modern tech stack and a self-hosted solution. ### Key Features: - **Agent-Native Infrastructure**: MCP server and REST API with full CRUD, schema access, activity history, and pipeline analysis. - **Customizable Data Model**: Custom fields for text, dates, currency, selects, and links between records, with per-field encryption. - **Multi-Workspace Isolation**: Workspace-scoped data with role-based permissions. - **Modern Tech Stack**: Laravel 13, Filament 5, PHP 8.5, with an automated test suite on every change. ### Why It Matters: Relaticle is perfect for teams looking to integrate AI agents without vendor lock-in, offering a developer-friendly, privacy-first solution. ### Tech Stack: - PHP 8.5 - PostgreSQL 17 - Composer 2 - Node.js 22+ - Redis for queues (optional for development) ### Installation: ```bash git clone https://github.com/Relaticle/relaticle.git cd relaticle && composer app-install ``` ### Development: ```bash composer dev composer test composer lint ``` ### Self-Hosting: See

### Overview: SymbolicAI is a neuro-symbolic framework that combines classical Python programming with the differentiable, programmable nature of Large Language Models (LLMs), making it easy to integrate semantic reasoning into your projects. ### Key Features: - **Neuro-symbolic Engine**: A modular framework that leverages the strengths of both classical Python and LLMs. - **Syntactic and Semantic Symbols**: `Symbol` objects that can be either syntactic or semantic, with semantic operations being the default. - **Extensibility**: Easily custom-engineered, hosted locally, or interfaced with various tools like web search and image generation. - **Documentation and Papers**: Comprehensive documentation and references to the associated paper and DeepWiki. ### Tech Stack: Python ### Why It Matters: SymbolicAI simplifies the integration of semantic reasoning into Python projects, making it easier to build projects that leverage the strengths of both classical programming and LLMs. ### Repository: ExtensityAI/symbolicai ### GitHub Stars: 1776 ### GitHub Description: A neuro-symbolic perspective on LLMs

Overview: This repository contains the PyTorch implementation of Fully Convolutional Networks, including training code to reproduce the original results. Key features: Training code, PyTorch implementation, Fully Convolutional Networks, TorchVision, Fcn, Pillow, Scipy, Tqdm. Tech stack: Python, PyTorch, TorchVision, Pillow, Scipy, Tqdm. Why it matters: This project provides a practical implementation of Fully Convolutional Networks using PyTorch, which can be used for various applications such as image segmentation tasks.

Overview: RAPTOR, the Recursive Abstractive Processing for Tree-Organized Retrieval, is an innovative approach to language modeling that constructs a recursive tree structure from documents to enhance information retrieval. This method addresses limitations in traditional language models by providing more efficient and context-aware retrieval across large text datasets. Key features: - Recursive tree structure from documents - Efficient and context-aware information retrieval - Addresses limitations in traditional language models Tech stack: Python Why it matters: RAPTOR introduces a novel method for language modeling that improves upon traditional approaches by enhancing retrieval efficiency and context-awareness, making it particularly useful for large-scale text datasets and complex information retrieval tasks. Repository: parthsarthi03/raptor GitHub stars: 1772 GitHub description: The official implementation of RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval. The repository includes the original paper, installation instructions, and basic usage examples.

### Overview: Explore a variety of deep learning models for speech synthesis and text-to-speech, seamlessly integrated into Google Colab. ### Key Features: - NVIDIA Tacotron 2 and Waveglow demo - NVIDIA Flowtron multi-speaker TTS demo - Rayhane-mamah Tacotron 2 demo - Griffin-Lim algorithm demo - Mongolian TTS demo - Pytorch DC-TTS demo - WaveRNN demo - Mozilla TTS demo - NVIDIA Mellotron demo - Real-Time Voice Cloning demo - ESPnet TTS notebook - ForwardTacotron LJSpeech TTS notebook ### Tech Stack: Jupyter Notebook ### Why It Matters: These models and demos provide a comprehensive overview of state-of-the-art deep learning techniques for speech synthesis and text-to-speech, making it easy for developers to experiment and understand these technologies.

Overview: AgentDock is a framework designed for building sophisticated AI agents with configurable determinism, providing a framework-agnostic and provider-independent backend-first approach. It includes an open-source client application and is built with TypeScript, focusing on simplicity, extensibility, and configurable determinism. Key features: - AgentDock Core: A framework for building and deploying AI agents. - Open Source Client: A Next.js application for reference implementation and consumer of the AgentDock Core framework. - TypeScript: Emphasizes simplicity, extensibility, and configurable determinism. - Comprehensive Prompt Library: A curated collection of prompts for various scenarios. Tech stack: TypeScript, AgentDock Core, Open Source Client. Why it matters: AgentDock simplifies the process of building AI agents, making it ideal for developers looking to create reliable and predictable systems that can operate with minimal supervision. The framework's focus on configurable determinism ensures that agents can be tailored to specific needs, enhancing their effectiveness and reliability.

### Overview: ModernBERT aims to bring BERT into modernity by introducing FlexBERT, a modular approach to encoder building blocks, and utilizing YAML configuration files for model building. This repository builds upon MosaicBERT, specifically the unmerged fork bringing Flash Attention 2, under the Apache 2.0 license. The project is focused on pre-training and evaluations, with a high-level overview and technical details available in the release blog post and arXiv preprint. ### Key Features: - FlexBERT: Modular encoder building blocks - YAML configuration files for model building - Extensive documentation and reproducibility - Supports both text and tokenized data - Uses MDS format for datasets, with decompressed support for MDS data ### Tech Stack: Python ### Why it Matters: This repository is essential for researchers and developers looking to modernize BERT and build more efficient and scalable models. It provides a framework for pre-training and evaluations, making it a valuable tool for AI research and development. Repository: AnswerDotAI/ModernBERT