Projects

### Overview A comprehensive library and textbook for agentic AI architectures, featuring 35 production-grade patterns that are end-to-end and provider-agnostic. Each architecture is equipped with a fully executed Jupyter notebook, real LLM outputs, and a benchmark leaderboard. ### Key Features - **Architecture Library**: 35 production-grade agentic AI patterns. - **End-to-End**: Real LLM outputs, provider-agnostic, deterministic-picker discipline. - **Benchmark Leaderboard**: 17 tasks, ranking every architecture against every relevant task. ### Tech Stack - Python library and runnable textbook. ### Why It Matters - **Provider-Agnostic**: Runs on multiple LLM providers (Nebius, OpenAI, Anthropic, etc.). - **Deterministic-Picker Pattern**: Central technical discipline, applied in 13 of 35 architectures. - **Real Outputs**: Captured run theory against synthetic examples, avoiding LLM-as-Scorer pathology. ### Repository - **Repository**: FareedKhan-dev/all-agentic-architectures - **GitHub Stars**: 4619 - **Primary Language**: Jupyter Notebook - **License**: MIT License - **Docs**: https://fareedkhan-dev.github.io/all-agentic-architectures/ ### Quickstart ```bash pip install "agentic-architectures[nebius,faiss,tavily]" ``` ```python from agentic_architectures import get_llm from agentic_architectures.architectures

Overview: Serenata de Amor is an open project using artificial intelligence to track government spending and make open data accessible for the public. Key features: - Rosie: An AI that analyzes congressional expenses and engages citizens in discussions about suspicious spendings. - Jarbas: A website that visualizes and interprets data generated by Rosie, enabling citizens to interact with politicians and state governments about public spending. Tech stack: Python, JavaScript, and other open-source technologies. Why it matters: Empowering citizens with data is crucial for fostering accountability and transparency in government spending, ensuring that citizens can participate in discussions about public finances and hold their representatives accountable. Repository: okfn-brasil/serenata-de-amor GitHub stars: 4617 GitHub description: 🕵 Artificial Intelligence for social control of public administration | **This repository does not receive frequent updates. Check out the README**

### Overview: Neuralangelo is an official implementation of "Neuralangelo: High-Fidelity Neural Surface Reconstruction," presented at the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2023. This project offers a comprehensive solution for high-fidelity neural surface reconstruction, leveraging the Imaginaire library from NVIDIA. ### Key features: - **High-Fidelity Surface Reconstruction**: Neuralangelo provides advanced neural surface reconstruction techniques for high-quality surface extraction. - **Collaborative Development**: The project includes contributions from multiple researchers, ensuring a robust and diverse set of features and functionalities. - **Versatile Usage**: The codebase supports both single-GPU and multi-GPU training, making it suitable for various computational environments. ### Tech stack: - **Python**: The primary language used for developing the project. - **Imaginaire Library**: Built upon the Imaginaire library from NVIDIA, providing a solid foundation for neural surface reconstruction. ### Why it matters: Neuralangelo offers significant improvements in surface reconstruction, enabling high-fidelity models that can be applied across various fields such as computer graphics, medical imaging, and more. Its contributions to the field of neural surface reconstruction are highly valuable and impactful. ### Repository: - **Repository**: NVlabs/neuralangelo - **GitHub Stars**: 4615 - **GitHub Description**: Official implementation of "Neuralangelo: High-Fidelity Neural Surface Reconstruction" (CVPR 2023) ### Installation: - **Docker Images**: - `docker.io/chenhsuanlin/colmap:3.

Overview: This repository contains tutorials for beginners to intermediate developers on using PyTorch and TorchText for sentiment analysis. It covers sequence classification models using neural bag-of-words, recurrent neural networks, convolutional neural networks, and transformers. Key features: - Sequence classification workflow with PyTorch - Neural bag-of-words model - Recurrent neural network (RNN) implementation - Convolutional neural networks (CNNs) for sentiment analysis - Transformers library for pre-trained BERT model Tech stack: PyTorch, TorchText, Jupyter Notebook Why it matters: This repository provides practical, step-by-step tutorials for developers to understand and implement sequence classification models using PyTorch and TorchText, making it easier for developers to apply sentiment analysis techniques to their projects.

### Overview: AutoTrain Advanced is a no-code solution designed for fast and easy training of state-of-the-art machine learning models. It simplifies the process of fine-tuning large language models (LLMs), object recognition models (ORPOs), dialogue policy optimization (DPOs), and reward models, among others. The project is aimed at developers and researchers looking to expedite the training process without the need for extensive coding. ### Key Features: - **LLM SFT Finetuning**: Fine-tuning of large language models (LLMs) using a Python notebook. - **LLM ORPO Finetuning**: Fine-tuning of ORPO models using a Python notebook. - **LLM DPO Finetuning**: Fine-tuning of dialogue policy optimization (DPO) models using a Python notebook. - **LLM Reward Finetuning**: Fine-tuning of reward models using a Python notebook. - **LLM Generic/Default Finetuning**: Fine-tuning of generic LLMs using a Python notebook. - **Text Classification**: Fine-tuning of text classification models using a Python notebook. - **Text Regression**: Fine-tuning of text regression models using a Python notebook. - **Token Classification**: Coming soon. - **Seq2Seq**: Coming soon. - **Extractive Question Answering**: Coming soon. - **Image Classification**: Coming soon. ### Tech Stack: AutoTrain Advanced uses Python for its core functionalities. ### Why It Matters: AutoTrain Advanced is designed to be user-friendly, allowing developers to quickly train advanced machine learning models without the

### Overview: PySOT is a software system designed by SenseTime Video Intelligence Research team, implementing state-of-the-art single object tracking algorithms like SiamRPN and SiamMask. It is written in Python and powered by the PyTorch deep learning framework, enabling high-quality and high-performance tracking research. ### Key Features: - Implements SiamRPN and SiamMask algorithms - Uses ResNet, MobileNetV2, and AlexNet backbone architectures - Provides a large set of baseline results and trained models in the PySOT Model Zoo - Supports evaluation on OTB2015, VOT16/18/19, VOT18-LT, LaSOT, and UAV123 datasets ### Tech Stack: PySOT is written in Python and utilizes the PyTorch deep learning framework. ### Why it Matters: PySOT is essential for researchers looking to implement and evaluate novel single object tracking algorithms, providing a high-quality, high-performance codebase that supports rapid implementation and evaluation of research projects.

### Overview: 《大语言模型》是一本专注于大语言模型技术的中文书籍，旨在为大模型初学者提供系统的技术讲解。本书内容深入结合了编者在研发大模型过程中的第一手经验，全面覆盖了大模型技术的多方面知识。本书适用于具有深度学习基础的读者阅读，可以作为一本基础的大模型参考书籍。 ### Key features: - 系统讲解大模型技术的多方面知识。 - 适用于具有深度学习基础的读者阅读。 - 作为一本基础的大模型参考书籍。 - 多个章节内容初稿多次修正，力求表达的清晰性与准确性。 ### Tech stack: 本书使用Python作为主要编程语言。 ### Why it matters: 本书是一部系统介绍大语言模型技术的中文图书，涵盖了预训练、微调、对齐、提示工程等众多基础内容，能够为相关从业人员提供权威的、系统的学习参考。

### Overview: TensorFlow Datasets is a collection of public datasets ready to use with TensorFlow and Jax, designed to simplify the process of working with machine learning models. ### Key features: - **Public Datasets**: TensorFlow Datasets includes a wide range of public datasets, such as MNIST, CIFAR-10, and more. - **Ease of Use**: The library is designed to be simple and straightforward, making it easy to get started with TensorFlow. - **Performance Optimization**: TFDS follows best practices for data performance, aiming for state-of-the-art speed. - **Reproducibility**: All users receive the same examples in the same order, ensuring reproducibility. - **Customizability**: Advanced users can have fine-grained control over the datasets. ### Tech stack: Python ### Why it matters: TensorFlow Datasets simplifies the process of working with machine learning models by providing a collection of public datasets ready to use with TensorFlow and Jax. It aims to make the integration of TensorFlow datasets as seamless as possible, focusing on simplicity, performance, and reproducibility.

### Overview: RuVector is a Rust-native substrate designed for fast, local decisions and agent memory. It combines local semantic embeddings, persistent vector retrieval, graph relationships, explicit feedback learning, memory lifecycle controls, and optional shared memory. Built by Reuven Cohen as part of the ruvnet open source AI stack, RuVector offers a comprehensive solution for enterprise-level AI applications. ### Key Features: - **Real-Time Decisions**: Fast, local decisions with low latency. - **Persistent Memory**: Stores context and evidence for recall. - **Graph Relationships**: Handles complex relationships between data points. - **Explicit Feedback Learning**: Adapts based on explicit feedback. - **Memory Lifecycle Controls**: Manages the lifecycle of memory and context. - **Optional Shared Memory**: Supports distributed systems with shared memory. ### Tech Stack: Rust ### Why It Matters: RuVector provides a powerful and efficient solution for real-time decision-making and memory management in AI applications. Its ability to handle complex relationships and adapt based on feedback makes it suitable for a wide range of enterprise-level AI systems, including those requiring high performance and real-time responsiveness. ### Repository: - **Repository**: [ruvnet/RuVector](https://github.com/ruvnet/RuVector) - **Stars**: 4,588 - **Description**: RuVector is a Rust-native substrate for fast local decisions and agent memory across sessions. It combines local semantic embeddings, persistent vector retrieval, graph relationships, explicit feedback learning, memory lifecycle controls, and optional shared memory. ### System Diagram: - **System

### Overview: Tree of Thoughts is an implementation of a large language model designed to reason through problems by proposing multiple steps and evaluating their effectiveness. It elevates model reasoning by at least 70% compared to traditional methods, making it a powerful tool for enhancing the performance of large language models in problem-solving tasks. ### Key features: - **Tree of Thoughts Model**: A large language model that proposes multiple next steps, scores them, and prunes weak branches to find the best solution. - **Model Proposals**: The model generates multiple candidate next steps from a given input, either all at once or one per step, and evaluates their effectiveness. - **Search Algorithms**: The model explores the problem space using breadth-first search (BFS) or depth-first search (DFS) with backtracking, pruning weak branches based on evaluation scores. - **Evaluation Criteria**: The model scores candidate steps on their ability to reach the target value, either individually or by comparing candidates and voting. - **Token Efficiency**: The model outputs are validated against a Pydantic schema, ensuring that the search does not parse free-form prose, making it highly efficient. ### Tech Stack: - **Python**: The primary programming language used for implementation and configuration. - **Swarms Framework**: A framework that integrates the Tree of Thoughts model into a larger system, providing a more comprehensive approach to problem-solving. ### Why it Matters: - **Enhanced Reasoning**: The Tree of Thoughts model significantly improves the reasoning capabilities of large language models, making them more effective at solving complex problems. - **Performance Improvement**: By proposing

### Overview: nono is a complete agent runtime security tool designed to secure your agents, their tools, and the skills they follow. It provides low-latency, minimal footprint, and zero setup protection with kernel-enforced sandboxing. ### Key features: - **Tools Sandbox**: Run each tool in its own ephemeral sandbox, isolated from the agent and other tools, with explicit permissions for files, network access, environment variables, and credentials. - **Credential Injection**: Keep API keys outside the agent’s sandbox with proxy-based injection and accessible to a single tool if need be. - **Network Access Control**: Control where agents connect and what they can do there, with domain filtering and API rules for HTTP methods and paths. - **Skill Trust**: Verify the origin and integrity of skills, hooks, and instruction files with Sigstore. Define which publishers you trust and reject unsigned or tampered files covered by your policy. - **Cryptographic Audit**: Keep a tamper-evident record of agent sessions and security decisions, with integrity verification and optional signing. - **Atomic Rollbacks**: Provide rollback and recovery capabilities to ensure security and integrity. ### Tech stack: Rust ### Why it matters: nono is a critical tool for securing agents and their environments, ensuring low latency, minimal footprint, and zero setup. It provides a comprehensive security solution for developers and organizations looking to enhance their security posture.

### Overview: RecBole is a comprehensive and efficient recommendation library developed in Python and PyTorch, designed to support research and development of recommendation algorithms. It includes 94 algorithms and 44 datasets, providing a unified and flexible data structure, benchmark models, and efficient GPU-accelerated execution. ### Key Features: - **General and Extensible Data Structure**: Unified and extensible data structures for various recommendation datasets. - **Comprehensive Benchmark Models and Datasets**: 94 commonly used recommendation algorithms and 44 formatted datasets. - **Efficient GPU-Accelerated Execution**: Optimized for GPU environments to enhance performance. - **Extensive and Standard Evaluation Protocols**: Supports widely adopted evaluation protocols for testing and comparing algorithms. ### Tech Stack: Python and PyTorch ### Why It Matters: RecBole provides a unified, efficient, and comprehensive framework for researchers and developers to work with recommendation algorithms, making it easier to reproduce and test various models and datasets.

Overview: Ceres Solver is a powerful, open-source C++ library designed for solving large-scale non-linear optimization problems. It is widely used in production at Google since 2010 and is known for its performance and feature-rich capabilities. Key features: - Solves non-linear Least Squares problems with bounds constraints - Solves general unconstrained optimization problems Tech stack: C++ Why it matters: Ceres Solver is essential for engineers and developers working on complex optimization tasks, providing a robust and efficient solution to large-scale optimization problems.

### Overview: Ultravox is a fast multimodal language model designed for real-time voice interactions, built on the latest advancements in multimodal understanding and natural language processing. ### Key Features: - **Real-Time Voice Response**: Ultravox converts audio directly into text, enabling immediate and seamless voice-to-voice interactions. - **Open-Weight Model Compatibility**: Compatible with various open-weight language models, including Llama 3, Mistral, and Gemma. - **Training**: Supports training on any open-weight model, available on Hugging Face. - **Demo**: Available on our demo page, where you can build your own voice-to-voice agents. - **Discord**: Join our Discord server for community support and discussions. - **Jobs**: Interested in working on Ultravox fulltime? Check out our jobs page. - **Inference Server**: Try out Ultravox using your own audio content with BaseTen's managed APIs. - **Model**: Download the latest weights from the Ultravox Hugging Face page. - **Architecture**: See the architecture diagram for details. ### Tech Stack: Python ### Why It Matters: Ultravox is designed to bridge the gap between text and speech, offering a more efficient and intuitive way to interact with language models. This model's ability to understand and respond to voice commands in real-time makes it particularly useful in applications requiring immediate and natural language processing. ### GitHub Stars: 4582 ### Latest News: - 2025/12 - [Ultravox 0.7](https

### Overview: 本项目为OpenClaw中文教程，包含安装、配置、实战案例和避坑指南，旨在帮助读者快速上手并掌握OpenClaw技术。 ### Key features: - 安装与配置指南 - 实战案例展示 - 避坑指南 - 作者的GPT-6/ Codex/ Claude Code新项目 ### Tech stack: Shell ### Why it matters: 本书全面覆盖了OpenClaw的安装、配置、实战案例和避坑指南，是学习OpenClaw的首选资源。

### Overview: ClearerVoice-Studio is an open-source AI-powered speech processing toolkit designed for researchers, developers, and end-users. It provides advanced capabilities such as speech enhancement, speech separation, and target speaker extraction. The toolkit includes state-of-the-art pre-trained models and comprehensive training and inference scripts. ### Key features: - **Speech Enhancement**: Enhances the quality of speech by improving its clarity and intelligibility. - **Speech Separation**: Separates multiple speakers in a mixed audio recording. - **Target Speaker Extraction**: Identifies and isolates specific speakers in a recorded conversation. - **Speech Super-Resolution**: Converts low-resolution audio to high-resolution, improving perceptual quality. ### Tech stack: Python ### Why it matters: ClearerVoice-Studio offers a powerful and versatile toolkit for researchers and developers looking to enhance speech processing capabilities. Its state-of-the-art models and comprehensive documentation make it an essential tool for anyone working in the field of speech processing and AI.

### Overview: Observal is a self-hosted registry for coding agent extensions, providing a built-in insight engine. It allows developers to easily share their Skills, MCPs, and Agents with peers. ### Key Features: - **Observal Registry**: A centralized platform for managing and sharing coding agent resources. - **Insight Engine**: Analyzes and provides insights into the usage and performance of agents. - **CLI Interface**: Observal CLI enables developers to interact with Observal through the command line. ### Tech Stack: - **Python**: Observal is built using the Python programming language. ### Why it Matters: Observal simplifies the management and sharing of coding agent resources within a team, ensuring that everyone has access to the latest tools and insights. This tool is particularly useful for tech-forward organizations looking to streamline their internal development processes.

Overview: This repository is a curated collection of tutorials and resources for Large Language Models, AI Painting, and related fields. Key features: - [CSCI-GA.3033-102 Special Topic - Learning with Large Language and Vision Models](https://www.sainingxie.com/llvm-fall23/) - [CS11-711 Advanced Natural Language Processing](https://phontron.com/class/anlp2024/) - [AI-Systems (LLM Edition) 294-162](https://learning-systems.notion.site/learning-systems/AI-Systems-LLM-Edition-294-162-Fall-2023-661887583bd340fa851e6a8da8e29abb) Tech stack: Unknown Why it matters: This repository provides a comprehensive collection of tutorials and resources for Large Language Models, AI Painting, and related fields, catering to both beginners and advanced AI enthusiasts.

### Overview: agent-memory is a long-term memory runtime designed for AI agents. It stores Markdown files as the single source of truth, providing local and ranked retrieval with an independent sleep-time management layer. The system uses Claude Code, Codex CLI, and Muse Code to share a single store, eliminating the need for API keys. ### Key Features: - **Retrieval Engine**: Handles embeddings, knowledge graph, and ranking pipelines. - **Local and Ranked Retrieval**: Answers with paths rather than pasted text, allowing agents to open only the necessary information. - **Filesystem**: Reads markdown files directly, providing a legible and cost-free filesystem. - **Write Coverage**: Writes are triggered at conversation boundaries, ensuring full trace coverage without holding up the task. ### Tech Stack: - Python 3.12 ### Why It Matters: agent-memory addresses the issue of agents forgetting what they learned in their sessions by providing a persistent, local, and ranked memory store for any AI agent, including those not just coding ones. It eliminates the need for API keys, ensuring privacy and security.

Overview: LongMemory is a local persistent memory store designed specifically for LLM applications, including Claude desktop, GitHub Copilot, Codex, and Antigravity. It offers a durable, immutable, and temporal memory solution that ensures the integrity and truth of data. Key features: - **Immutable Content**: Ensures that data is not modified after ingestion. - **Temporal Truth**: Provides a historical and associative recall mechanism. - **Governed Project Memory**: Supports Skills, Chat Memory, LLM-Wiki, and CodeGraph. - **One TypeScript Engine**: Cross-platform support for npm, CLI, HTTP, MCP, and VS Code. - **Native Integrations**: Seamless integration with various agent hosts, automation tools, and Python frameworks. Tech stack: TypeScript. Why it matters: LongMemory is essential for applications that require a reliable and explainable memory solution, ensuring that models remain stateless while maintaining the integrity of the application's data. Repository: CaviraOSS/LongMemory GitHub stars: 4541 GitHub description: Local persistent memory store for LLM applications including Claude desktop, GitHub Copilot, Codex, and Antigravity.