Projects

### Overview: FinGLM是一个致力于构建一个开放、公益和持久的金融大模型项目，利用开源开放来促进“AI+金融”。项目通过赛题挑战赛，吸引了来自多方机构的参与，旨在探索AI在金融领域的边界。 ### Key features: - **赛题挑战赛**：FinGLM通过赛题挑战赛，吸引了来自多方机构的参与，探索AI在金融领域的边界。 - **开源数据**：FinGLM开源了包括年报数据集、标注数据等在内的大量数据，为后续的研究和应用提供了基础。 - **模型微调**：项目提供了一整套微调流程，包括数据准备、模型准备和问答流程，确保模型的持续优化和应用。 - **开源策略**：FinGLM项目计划将所有团队的方案、代码和模型开源，提供持续的迭代和优化。 ### Tech stack: HTML ### Why it matters: FinGLM项目通过开源和开放的策略，为AI在金融领域的应用提供了基础和方向，推动了AI技术在金融领域的深入发展和应用。

Overview: `stable-worldmodel` is a platform designed for reproducible research in world model evaluation, offering a unified interface for collecting data, training, and evaluating models across various environments. Key features: - Data collection with `World` class - Training with `WorldModelPolicy` and `PlanConfig` - Evaluation using `CEMSolver` - Support for multiple environments and datasets Tech stack: Python Why it matters: This platform simplifies the process of world model research by providing a standardized way to collect, train, and evaluate models, making it easier for researchers to focus on the core contributions of their work. Repository: galilai-group/stable-worldmodel GitHub stars: 2267 GitHub description: A platform for reproducible world model research and evaluation

Overview: SageMaker Python SDK is a library designed for training and deploying machine learning models on Amazon SageMaker, offering support for popular deep learning frameworks like MXNet and PyTorch, as well as Amazon algorithms optimized for GPU training. Key features: - Training and deploying models using MXNet and PyTorch - Training and deploying models with Amazon algorithms - Training and hosting models using your own algorithms built into SageMaker compatible Docker containers Tech stack: Python Why it matters: SageMaker Python SDK simplifies the process of training and deploying machine learning models on Amazon SageMaker, making it easier for developers to build, train, and deploy models efficiently.

Overview: MemoRAG is an innovative RAG framework built on a highly efficient, super-long memory model, designed to enhance evidence retrieval and generate contextually rich responses. Unlike standard RAG, it leverages its memory model to achieve a global understanding of the entire database. Key features: - Global Memory: Handles up to 1 million tokens in a single context, providing comprehensive understanding across massive datasets. - Optimizable & Flexible: Adapts to new tasks with ease, achieving optimized performance with just a few hours of additional training. - Contextual Clues: Generates precise clues from global memory, bridging raw input to answers and unlocking hidden insights from complex data. - Efficient Caching: Speeds up context pre-filling by up to 30x, with support for caching chunking, indexing, and encoding. Tech stack: Python Why it matters: MemoRAG is a significant advancement in RAG technology, offering a more comprehensive and contextually rich approach to evidence retrieval, making it particularly useful for applications requiring global understanding and precise clue extraction from complex data.

### Overview: This repository is a comprehensive list of machine learning libraries written in Rust. It includes a variety of libraries and algorithms, categorized into basic libraries, algorithm categories, and small libraries. The repository also contains libraries that are no longer maintained and small libraries. It has commented on helpful parts of the code and good libraries within each category, making it a valuable resource for those considering migrating from Python to Rust for machine learning. ### Key Features: - **Basic Libraries**: Various libraries for common machine learning tasks. - **Algorithm Categories**: Categorization of machine learning algorithms. - **Small Libraries**: Libraries that are no longer maintained. - **Commented Code**: Helpful parts of the code and good libraries within each category. - **Migrating from Python**: Targeted at people thinking of migrating from Python to Rust for machine learning. ### Tech Stack: Rust ### Why It Matters: This repository is a valuable resource for anyone considering migrating from Python to Rust for machine learning. It provides a comprehensive list of libraries, categorized and commented, making it easier to find and understand the best tools for their needs. ### Repository: vaaaaanquish/Awesome-Rust-MachineLearning - **Primary Language**: JavaScript - **GitHub Stars**: 2267 - **GitHub Description**: This repository is a list of machine learning libraries written in Rust. It's a compilation of GitHub repositories, blogs, books, movies, discussions, papers, etc. - **GitHub Image**: ![arml](https://github.com/vaaaaanquish/Awesome-Rust

### Overview: ArcKit is a comprehensive toolkit designed for enterprise architects, transforming their governance workflows into a structured, AI-assisted process. It includes features for establishing principles, analyzing stakeholders, managing risks, creating requirements, and more, all with the goal of improving enterprise architecture governance. ### Key Features: - **Establishing and Enforcing Architecture Principles**: Ensures that architecture principles are adhered to. - **Analyzing Stakeholder Drivers, Goals, and Outcomes**: Helps in understanding and aligning with business goals. - **Risk Management (HM Treasury Orange Book)**: Manages risks using the Orange Book standards. - **Business Case Justification (HM Treasury Green Book SOBC)**: Provides justification for business cases. - **Creating Comprehensive Requirements Documents**: Ensures that requirements are clear and comprehensive. - **Data Modeling with ERD, GDPR Compliance, and Data Governance**: Supports data modeling and compliance. - **Technology Research with Build vs Buy Analysis**: Helps in technology research and analysis. - **Azure-specific Research Using Microsoft Learn MCP**: Provides authoritative documentation for Azure. - **Strategic Planning with Wardley Mapping**: Facilitates strategic planning and mapping. - **Generating Visual Architecture Diagrams (Mermaid)**: Creates visual representations of architecture. - **Managing Vendor RFP and Selection Processes**: Ensures vendor RFPs are managed effectively. - **Conducting Formal Design Reviews (HLD/DLD)**: Facilitates formal design reviews. - **Maintaining Requirements Traceability**: Ensures requirements are traceable. - **Citation Traceability for External Documents**: Provides trace

### Overview: This curated list contains Python packages for time series analysis, covering a wide range of techniques and applications. It serves as a valuable resource for developers looking to enhance their time series data processing and analysis capabilities. ### Key Features: - **Time Series Forecasting**: Libraries like Darts, ETNA, and Prophet are used for forecasting and analysis. - **Feature Extraction**: Tools such as Featuretools and khiva-python are designed for extracting features from time series data. - **Machine Learning Models**: Packages like glm-sklearn and pyDSE are used for machine learning applications. - **Python Libraries**: A variety of Python libraries are included, such as Arrow, Darts, and PyTorch, which are essential for time series analysis. ### Tech Stack: Python ### Why it Matters: Time series analysis is crucial in various fields such as finance, economics, and engineering. This list provides developers with the necessary tools to handle time series data effectively, enabling them to make informed decisions and predictions based on historical patterns and trends.

Overview: This project focuses on Chinese Named Entity Recognition (NER) and Relation Extraction (RE) using advanced deep learning techniques, specifically IDCNN/biLSTM+CRF for NER and biGRU+2ATT for RE. Key features: - Chinese Named Entity Recognition with IDCNN/biLSTM+CRF - Chinese Relation Extraction with biGRU+2ATT Tech stack: Python Why it matters: This project showcases the application of state-of-the-art deep learning methods in Chinese information extraction, making it valuable for researchers and developers looking to improve their NER and RE capabilities in Chinese language. Repository: crownpku/Information-Extraction-Chinese Primary language: Python GitHub stars: 2264

Overview: This repository contains source files for guides and tutorials on the Generative AI developer site, focusing on the Gemini API and Gemma. Key features: - Generative AI developer site guides and tutorials - Gemini API and Gemma documentation - Demos apps - Examples for specific concepts Tech stack: Jupyter Notebook Why it matters: These resources provide essential information and examples for developers working with the Gemini API and Gemma, helping them understand and implement generative AI solutions. Repository: google/generative-ai-docs Primary language: Jupyter Notebook GitHub stars: 2263 GitHub description: This repository is deprecated and will be archived

Overview: Agent Router is a unified access platform for managing and controlling access to various generative AI services, including OpenAI-compatible APIs, hosted providers, and self-hosted inference and model servers. It uses Envoy Gateway for routing and is an open-source project by the Agentic AI Foundation. Key features: - Unified access to multiple AI services - Centralized management of credentials, routing, quotas, failover, and usage attribution - Supports OpenAI, Azure OpenAI, and Google Gemini providers - CLI and Kubernetes deployment options Tech stack: Go Why it matters: Agent Router simplifies the management of OpenAI-compatible AI services, making it easier for developers to integrate with various AI providers and self-hosted model servers, while maintaining a consistent and secure API experience. Repository: theagentrouter/agent-router GitHub stars: 2257 GitHub description: Manages Unified Access to Generative AI Services built on Envoy Gateway

Overview: Turing.jl is a powerful probabilistic programming package for Bayesian inference, designed to seamlessly integrate with Julia's scientific computing ecosystem. It supports a variety of inference methods including MCMC, variational inference, and maximum likelihood estimation, making it highly versatile for Bayesian analysis. Key features: - Supports Julia 1.10.8 or later - Implements Bayesian and likelihood-based inference - Integrates seamlessly with Julia's scientific computing ecosystem - Provides models written in `@model`, JuliaBUGS syntax, and DoodlePPL - Offers automatic differentiation through ForwardDiff.jl and Mooncake.jl - Facilitates gradient-based algorithms with other backends like Enzyme.jl Tech stack: Julia Why it matters: Turing.jl simplifies Bayesian inference by providing a user-friendly interface and efficient computation, making it an essential tool for researchers and practitioners in various fields requiring probabilistic modeling and inference.

Overview: Forge is a Python framework designed for self-hosted Large Language Models (LLMs) that allows users to call their tools in any order without worrying about reliability or guardrails. It simplifies workflow management by providing a reliable and secure environment for executing complex multi-step processes. Key features: - **Guardrails**: Forge applies reliability and security measures to ensure that the model's responses are validated and that required steps are followed. - **Workflow Structure**: Users can define prerequisites, required steps, and terminal tools to control the workflow structure. - **Supports Multiple Backends**: Forge supports various backend models such as OpenAI, Anthropic, Ollama, llama-server (llama.cpp), Llamafile, vLLM, and Anthropic. Tech stack: Python Why it matters: Forge simplifies the process of integrating and managing LLMs, making it easier to build and deploy reliable, secure, and efficient workflows without the need for extensive coding or integration with existing agents. Repository: antoinezambelli/forge GitHub stars: 2261 GitHub description: A Python framework for self-hosted LLM tool-calling and multi-step agentic workflows

Overview: TigerBot is a cutting-edge language model designed for a wide range of tasks, including medical NER, search, and game development. It supports multiple languages and is trained on high-quality data to ensure accuracy and efficiency. Key features: - Supports multiple languages - Medical Named Entity Recognition (NER) model - Search function - Game development using large language models - High-quality training data for comprehensive results Tech stack: Python Why it matters: TigerBot is a significant advancement in the field of language models, offering a versatile platform for various applications and ensuring high performance across different languages and tasks. Repository: TigerResearch/TigerBot GitHub stars: 2260 GitHub description: TigerBot: A multi-language multi-task LLM

### Overview: JustHireMe is a local-first AI job intelligence workbench designed for scraping roles, ranking fit, and generating tailored application materials. It aims to provide a better alternative to noisy job boards and cloud-based AI apply tools. ### Key Features: - **Local-first Desktop Workbench**: The primary focus is on a local-first desktop workbench for efficient job intelligence. - **Python Sidecar API**: A sidecar API for Python-based applications, supporting various functionalities. - **Scraping and Ranking**: Capable of scraping roles and ranking fit based on specific criteria. - **Document Generation**: Generates tailored application materials to help candidates stand out. ### Tech Stack: JustHireMe is built using Python, with a focus on local-first architecture for better performance and reliability. ### Why It Matters: JustHireMe offers a more personalized and efficient job application process, making it a valuable tool for candidates looking to stand out in a competitive job market.

### Overview: This repository is a curated list of open-source and source-available tools designed for orchestrating agents, covering various functionalities such as parallel coding, autonomous runners, multi-agent swarms, personal assistants, and agent infrastructure. ### Key Features: - **Parallel Coding Agents — Terminal (TUI/CLI):** Supervises multiple agent sessions side by side from a terminal, including tmux panes, git worktrees, and TUI dashboards. - **Parallel Coding Agents — Desktop & Web:** Covers sessions across multiple agents like Claude Code, Codex, Gemini, and OpenCode with live status and resume for each. - **Multi-Agent Swarms:** Orchestrates agents to split large jobs among themselves. - **Autonomous Loop Runners:** Drives one goal until it verifies. - **Autonomous Task Runners:** Pulls work from an issue tracker, board, or schedule. - **Agent Infrastructure & Primitives:** Controls planes, coordination protocols, harness adapters, and runtimes. - **Personal Assistants:** Stay running, remember across sessions, and pick up work from a chat thread. - **Resting:** Orchestrates agents to work while you're away. ### Tech Stack: Rust, Python, JavaScript, and various terminal-based tools. ### Why it Matters: These tools enable efficient and effective management of agents, making it easier to manage and coordinate complex tasks across multiple sessions, ensuring that agents can work in parallel, run autonomously, and integrate seamlessly with other systems.

### Overview: This repository is an extensive collection of resources related to LLM-powered agents, including papers, repositories, and more. The goal is to provide a comprehensive overview of the latest advancements in this field, focusing on papers, open-source projects, and other relevant sources. ### Key Features: - Papers: A curated list of papers with over 50 citations or repositories with over 200 stars. - Papers (continued): Additional papers accepted by reputable conferences and journals. - Open-Source Projects: Repositories dedicated to autonomous task solvers, multi-agent simulations, and more. - Perspectives: Insights and discussions from experts in the field. - Other Related Sources: Additional resources and links to keep the project updated. ### Tech Stack: The project uses Markdown, GitHub, and other open-source tools for documentation and collaboration. ### Why it Matters: This repository aims to provide a valuable resource for researchers, developers, and enthusiasts interested in the latest advancements in LLM-powered agents. By keeping the content updated, it ensures that the latest developments in the field are accessible and useful.

### Overview: MT-DNN is a powerful framework for natural language understanding, implemented in PyTorch and designed for multi-task deep neural networks. This package addresses various language tasks, including language modeling, sequence labeling, and commonsense reasoning, among others. ### Key Features: - **Multi-Task Learning (MT-LM):** Fine-tuning pre-trained language models for various language understanding tasks. - **Knowledge Distillation:** Leveraging knowledge distillation to improve the robustness and generalization of pre-trained models. - **Hybrid Neural Networks:** Combining different neural network architectures to enhance performance in specific language tasks. ### Tech Stack: Python ### Why It Matters: MT-DNN provides a comprehensive solution for natural language understanding tasks, improving the performance and generalization capabilities of deep learning models. Its ability to handle multiple tasks efficiently makes it highly valuable in the field of language processing and artificial intelligence. Repository: namisan/mt-dnn GitHub Stars: 2258 GitHub Description: Multi-Task Deep Neural Networks for Natural Language Understanding

Overview: TensorFlow for Raspberry Pi is a project that provides official Python wheels for TensorFlow, making it officially supported for use on Raspberry Pi. Key features: - Officially supported TensorFlow wheels for Raspberry Pi Tech stack: Python Why it matters: TensorFlow for Raspberry Pi is now officially supported, allowing users to leverage the latest TensorFlow features on their Raspberry Pi without the need for additional dependencies.

### Overview: DSH Pocket 是一款将 DeepSeek Harness 装进口袋的应用，让你无论在何处都能实时访问和操控电脑上的 DeepSeek Harness 界面。 ### Key features: - **局域网扫码**: 设置 → 手机访问，打开就有局域网二维码，手机连同一 WiFi 扫码即开（自动识别本机局域网 IP，WSL 环境自动取 Windows 物理网卡 IP）。 - **局域网开关**: 设置页可一键关闭/开启局域网访问（切换时弹窗提醒）。 - **公网扫码**: 点「开启公网访问」→ cloudflared 隧道 → 出公网二维码，4G/任何网络都能访问。 - **公网固定域名**: 可选「命名隧道」模式：填 Cloudflare Tunnel Token + 自己的域名，公网地址固定不变（重启不再变；见下方说明）。 - **公网密码**: 公网链接默认使用 8 位随机密码（每次开启公网自动换新；可自定义固定密码）；局域网有独立的默认 8 位随机密码（默认开启，设置页可一键关闭——关闭后局域网扫码直连）。 - **会话保持**: 手机输一次密码后长期免输（登录状态绑定电脑上的 dsh web 进程：只要它不重启，手机不用再输；dsh web 重启/更新后需重新输入一次）。 - **实时同步

### Overview: Harmonist is a portable AI agent orchestration framework that enforces mechanical protocol checks on code changes, ensuring that every agent adheres to strict rules and protocols. It integrates seamlessly with various AI coding assistants and is designed to be used with Python 3.9+. ### Key features: - **Protocol Enforcement**: Every code change is gated by hooks that check for necessary approvals, memory updates, and supply chain integrity. - **193 Agents**: The framework supports 193 agents, including Cursor, Claude Code, Copilot, Windsurf, Aider, and others. - **Zero Runtime Dependencies**: Harmonist requires no runtime dependencies, making it easy to integrate into existing projects. ### Tech stack: - Python 3.9+ ### Why it matters: Harmonist is the first open-source agent framework that enforces protocol checks as mechanical gates, not polite requests in prompts. This ensures that AI coding assistants follow structural rules and protocols, preventing silent bugs and maintaining the integrity of the development process. ### Repository: Repository: **GammaLabTechnologies/harmonist** GitHub stars: **2255** GitHub description: Portable AI agent orchestration with mechanical protocol enforcement. 186 agents, zero runtime dependencies.