Projects

### Overview: This comprehensive guide offers a thorough review of AI system design, including RAG architectures, LLM engineering, agentic AI, MCP and A2A protocols, and interview preparation. It is designed to help engineers build and evaluate production AI systems. ### Key features: - **Interview Preparation**: Provides a 147-question interview bank and answer frameworks. - **RAG Fundamentals**: Covers retrieval-based architectures. - **LLM Engineering**: Discusses engineering aspects of large language models. - **Agentic AI**: Explains the concept of agentic AI. - **MCP and A2A Protocols**: Details the Multiple Choice Protocol and Alternative Answer Protocol. - **Production Patterns**: Offers insights into production patterns for AI systems. ### Tech stack: Python, Markdown, HTML, CSS, and other relevant technologies. ### Why it matters: This guide is essential for engineers building and evaluating production AI systems, providing a deep dive into the practical aspects of AI engineering and interview preparation. Repository: ombharatiya/ai-system-design-guide

### Overview: 这个项目是百度官方出品的一站式深度学习在线百科，涵盖了从入门到进阶的深度学习课程、书籍、案例和面试题库，旨在帮助开发者高效学习和掌握深度学习知识。 ### Key features: - 零基础实践深度学习课程 - 产业实践深度学习课程 - 特色课程 - 飞桨套件课程汇总资料 - 《动手学深度学习》书籍 - 深度学习百问 - 面试宝典 - 飞桨产业实践范例库 - 飞桨各产品课程 ### Tech stack: Jupyter Notebook ### Why it matters: 本项目内容全面、形式丰富、实时更新，提供从理论到实践、从科研到产业应用的深度学习学习材料，旨在帮助开发者高效学习和掌握深度学习知识，快速成为AI跨界人才。

### Overview: ReMe is a memory management kit designed for agents, enabling them to remember, refine, and evolve knowledge. It transforms conversations and resources into readable, editable, and searchable Markdown, allowing agents and users to interact with the knowledge in familiar ways. ### Key Features: - **Memory as File, File as Memory**: ReMe stores durable memory as ordinary Markdown with frontmatter and wikilinks, making it easy to inspect, edit, move, sync, and back up with familiar tools. - **Self-evolving knowledge base**: ReMe progressively turns conversations and resources into daily notes and long-term knowledge, preserving sources while refining facts, preferences, procedures, and relationships over time. - **Recall is precise and context-aware**: BM25, optional embeddings, and wikilink expansion retrieve relevant line-level passages and their relationships without loading the entire knowledge base into the agent context. - **One memory workspace works across agents**: Personal assistants, coding agents, and other agent runtimes can share the same local workspace through native integrations, SKILL.md, CLI, HTTP, MCP, or Python APIs. ### Tech Stack: - Python ### Why It Matters: ReMe is a crucial tool for agents and users alike, providing a seamless and efficient way to manage and evolve knowledge. It allows agents to collaborate and share knowledge in a way that is both intuitive and powerful, making it an indispensable part of the agent ecosystem.

Overview: SimCSE is a simple contrastive learning framework designed for both unlabeled and labeled data, providing state-of-the-art sentence embeddings. It was presented at EMNLP 2021 and has been accepted to the conference. Key features: - Unsupervised SimCSE: Predicts input sentences in a contrastive learning framework with standard dropout. - Supervised SimCSE: Incorporates annotated pairs from NLI datasets using entailment and contradiction as positives and negatives. - Sentence embedding tool: Easy-to-use package for encoding sentences into embeddings and computing cosine similarities between groups of sentences. - Model list: Full list of available models, including both unsupervised and supervised versions. - Training code: Provides requirements, evaluation, and training instructions. - Evaluation code: Available for model checkpoints and evaluation. - Model checkpoints: Pre-trained models for out-of-the-box usage. - Demo code: Available for training and evaluation. - Paper citation: Refer to the accepted paper for further details. Tech stack: Python Why it matters: SimCSE offers a simple yet effective solution for sentence embedding, making it a valuable tool for natural language processing tasks.

Overview: Dataherald is a natural language-to-SQL engine designed for enterprise-level question answering over relational data. Key features: - Allows business users to get insights from the data warehouse without going through a data analyst. - Enables Q+A from production DBs inside SaaS applications. - Creates a ChatGPT plug-in from proprietary data. Tech stack: Python Why it matters: Dataherald simplifies data querying by translating natural language into SQL, making it accessible to non-technical users while maintaining the power of enterprise-level data management.

Overview: Rio is a Python-based framework for building websites and applications, eliminating the need for HTML, CSS, and JavaScript. Key features: - No need for HTML, CSS, or JavaScript - Built-in React-style components - Rich component library - Seamless integration with web and local environments Tech stack: Python Why it matters: Rio simplifies web development by eliminating the need for traditional frontend technologies, making it easier to create modern, user-friendly applications without the complexity of JavaScript and HTML.

### Overview: HolmesGPT is a CNCF SRE Agent designed for investigating production incidents and finding root causes across various stacks, including Kubernetes, VMs, cloud providers, databases, and SaaS platforms. It runs in the background, spotting problems proactively and providing notifications via Slack. ### Key Features: - **Operator Mode**: Runs in the background 24/7, automatically fixing issues before customers notice. - **Petabyte-scale Data**: Server-side filtering, JSON tree traversal, and tool output transformers keep large payloads out of context windows. - **Memory-safe Execution**: Per-tool memory limits, streaming large results to disk, and automatic output budgeting prevent OOM kills. - **Deep Integrations**: Built-in with Prometheus, Grafana, Datadog, Kubernetes, and many more REST APIs. - **Bidirectional Alert Integrations**: Fetch alerts from AlertManager, PagerDuty, OpsGenie, or Jira—and write findings back. - **Any LLM Provider**: Supports OpenAI, Anthropic, Azure, Bedrock, Gemini, and more. - **No Kubernetes Required**: Works with any infrastructure—VMs, bare metal, cloud services, or containers. ### Tech Stack: Python ### Why It Matters: HolmesGPT is a CNCF SRE Agent designed for investigating production incidents and finding root causes across various stacks, including Kubernetes, VMs, cloud providers, databases, and SaaS platforms. It runs in the background, spotting problems proactively and providing notifications via Slack. ### Repository: HolmesGPT/holmesgpt ###

### Overview: A personal knowledge base that continuously grows and improves itself, built and maintained by the agent. It reads and extracts knowledge from various sources, creating a structured wiki that compiles and synthesizes information over time. ### Key Features: - **Persistent Wiki**: A structured markdown-based knowledge base that accumulates across sessions. - **Entity Pages**: Auto-created pages for every person, company, or project mentioned in sources, updated each time a new source references them. - **Concept Pages**: Auto-created pages for every key idea, framework, or method mentioned in sources. - **Graph Visualization**: An interactive graph visualization of the knowledge base, built from all wikilinks. ### Tech Stack: - Python ### Why It Matters: This knowledge base allows for a more structured and comprehensive way of storing and accessing information, making it easier to find and synthesize knowledge across different sources. It eliminates the need for manual note-taking and ensures that all sources are integrated and cross-referenced, providing a richer and more interconnected knowledge base.

Overview: NExT-GPT is a state-of-the-art multimodal language model designed for the ICML 2024 paper. It is an Any-to-Any multimodal large language model that can generate text, images, videos, and audio in arbitrary combinations. Key features: - Multimodal encoding stage using established encoders to project input representations into language-like form. - Harnessing an existing open-source LLM to interpret input information and generate text tokens. - Multimodal generation stage that maps signal token representations into understandable forms for following multimodal decoders. Tech stack: Python Why it matters: NExT-GPT is a significant advancement in the field of multimodal language models, offering the ability to generate outputs in various combinations of text, images, videos, and audio, making it highly versatile and useful in applications requiring multimodal interaction. Repository: NExT-GPT/NExT-GPT GitHub stars: 3635 GitHub description: Code and models for ICML 2024 paper, NExT-GPT: Any-to-Any Multimodal Large Language Model

Overview: ArtLine is a project aimed at creating line art portraits using deep learning techniques. Key features: - Line art portraits generation - ControlNet integration for enhanced portrait styles - Support for various image styles and poses Tech stack: Jupyter Notebook Why it matters: ArtLine leverages advanced deep learning models to produce highly detailed and realistic line art portraits, making it a valuable tool for artists and enthusiasts alike. Repository: vijishmadhavan/ArtLine GitHub stars: 3633 GitHub description: A Deep Learning based project for creating line art portraits.

### Overview: OpenDeepWiki is an open-source project aimed at providing a powerful knowledge management and collaboration platform. It leverages C# and TypeScript, focusing on modular design and ease of expansion. The project supports various features such as repository documentation, chat integration, and more. ### Key Features: - Import repository sources from Git URLs, ZIP archives, or approved local directories. - Generate README summaries, project overviews, wiki catalogs, document content, multi-language translations, mind maps, and Graphify artifacts. - Publish public repository docs on SEO-friendly routes, including `/{owner}/{repo}`, `/{owner}/{repo}/mindmap`, and `/{owner}/{repo}/graphify`. - Expose repository knowledge through repository-scoped MCP endpoints, the built-in chat assistant, embedded chat APIs, and share links. - Manage repositories, users, roles, departments, API keys, AI providers/models, skills, MCP providers, and GitHub App imports from the admin console. - Run background workers for repository processing, translation, mind map generation, Graphify artifacts, and scheduled incremental updates. - Support incoming chat/webhook channels for Feishu, QQ, WeChat, and Slack. ### Tech Stack: - C# - TypeScript ### Why it Matters: OpenDeepWiki simplifies the process of converting Git repositories, ZIP archives, and local directories into searchable knowledge bases. It generates structured repository documentation, serves it through a public Next.js site, and reuses the same indexed content for chat, embed, and MCP workflows. This tool is designed to be easy to expand and customize, making

Overview: LandPPT 是一个基于LLM的演示文稿生成平台，能够自动将文档内容转换为专业的PPT演示文稿。平台支持多种AI模型，提供丰富的模板和样式选择，让用户能够创建高质量的演示文稿。 Key features: - 一键生成 - 快速模式 - 智能配图 - 深度研究 - 讲稿与视频 - 多格式导出 Tech stack: Python Why it matters: LandPPT 是一个基于大语言模型的智能演示文稿平台，能够自动将文档内容转换为专业的PPT演示文稿，支持多种AI模型和丰富的模板，提供高质量的演示文稿。

### Overview: OpenArm is an open-source 7DOF humanoid arm designed for physical AI research and deployment in contact-rich environments. Built with high backdrivability and compliance, it is safe for human-robot interaction and practical for real-world applications. At $6,500 USD, it provides a flexible platform for teleoperation, imitation learning, simulation, and data collection in contact-rich tasks. ### Key Features: - **7DOF Arm**: High precision and flexibility for various research and practical applications. - **Safe Design**: Built for safe human-robot interaction, ensuring safety and compliance. - **Practical Payload**: Capable of handling practical payloads, making it suitable for real-world applications. - **Standardized Environment**: OpenArm Cell provides consistent evaluation conditions, facilitating global discussion on state-of-the-art physical AI research. ### Tech Stack: - **OpenArm Hardware**: CAD, STL, and BOM for OpenArm, Cell, and KER. - **OpenArm Description**: URDF/xacro robot description. - **OpenArm CAN**: CAN motor control library and CLI. - **OpenArm ROS2**: ros2_control and MoveIt packages. - **OpenArm MuJoCo**: MuJoCo models, including the Cell scene. - **OpenArm Isaac Lab**: Isaac Lab environments and RL tasks. - **OpenArm Dataset**: Dataset format, Python API, LeRobot conversion. - **OpenArm Teleop**: Teleoperation interface. ### Why It Matters: OpenArm is a crucial tool for researchers and developers looking to advance the field of physical AI

### Overview: Ploomber is the fastest way to build and deploy data pipelines, allowing you to develop and deploy applications without changing code. ### Key features: - A simple YAML API for quick start - Flexible Python API for total flexibility - Automatically caches previous results to speed up development - Deploy anywhere with Kubernetes, Airflow, AWS Batch, and SLURM ### Tech stack: Python ### Why it matters: Ploomber streamlines the process of building and deploying data pipelines, making it easier and faster to develop and deploy applications without changing the code.

Overview: PyBroker is a Python framework designed for developing algorithmic trading strategies, focusing on strategies that utilize machine learning. It provides a fast backtesting engine, easy rule creation and model integration, support for multiple time intervals, and comprehensive tools for model training and optimization. Key features: - A super-fast backtesting engine built in NumPy and accelerated with Numba - Easy creation of trading rules and models for executing across multiple instruments - Integration of trading signals across multiple time intervals (daily, weekly, monthly) - Access to historical data from Alpaca, Yahoo Finance, AKShare, or custom data sources - Model training and backtesting using Walkforward Analysis - Reliable trading metrics using bootstrapping for more accurate results - Parameter optimization with Optuna - Caching of downloaded data, indicators, and models for faster development - Parallelized computation and training for faster performance - Agent Skills that help AI agents write trading strategies and backtest using PyBroker Tech stack: Python, NumPy, Numba, Optuna, Walkforward Analysis, Bootstrapping, Alpaca, Yahoo Finance, AKShare, Custom Data Source, Parallelization, Caching, Agent Skills Why it matters: PyBroker simplifies the process of creating and evaluating algorithmic trading strategies that incorporate machine learning, making it easier to develop and test complex trading rules and models.

### Overview: Gemini-API is a reverse-engineered asynchronous Python wrapper for the Google Gemini web app, formerly known as Bard. It provides a comprehensive set of features including persistent cookies, image and video/audio generation, deep research, system prompt customization, extension support, classified outputs, streaming mode, and a standalone CLI tool. The official flavor is inspired by Google Generative AI's official API, offering a simple and elegant interface. ### Key Features: - **Persistent Cookies**: Automatically refreshes cookies in the background for always-on services. - **Image Generation**: Natively supports generating and editing images with natural language. - **Video & Audio Generation**: Supports generating videos and audio content natively. - **Deep Research**: Full deep research workflow with plan creation, status polling, and result retrieval. - **System Prompt**: Customizes the model's system prompt with Gemini Gems. - **Extension Support**: Supports generating content with Gemini extensions, such as YouTube and Gmail. - **Classified Outputs**: Categorizes text, thoughts, images, videos, and audio in the response. - **Streaming Mode**: Supports stream generation, yielding partial outputs as they are generated. - **CLI Tool**: Standalone command-line interface for quick interactions. - **Official Flavor**: Inspired by Google Generative AI's official API, offering a simple and elegant interface. - **Asynchronous**: Utilizes `asyncio` to run generation tasks and return outputs efficiently. ### Tech Stack: - Python ### Why It Matters: Gemini-API simplifies the process of interacting with the Google Gemini web app, making it easier

### Overview: This repository contains an extensive list of Python libraries and tools for data science, focusing on machine learning, deep learning, and related technologies. It is a curated collection that aims to provide the best resources for Python developers working in the field of data science. ### Key Features: - **Machine Learning Libraries**: - **SciPy**: Fundamental algorithms for scientific computing in Python. - **scikit-learn**: Machine learning in Python. - **PyCaret**: An open-source, low-code machine learning library in Python. - **Shogun**: Machine learning toolbox. - **xLearn**: High Performance, Easy-to-use, and Scalable Machine Learning Package. - **cuML**: RAPIDS Machine Learning Library, GPU-accelerated. - **modAL**: Modular active learning framework for Python3. - **Sparkit-learn**: PySpark + scikit-learn. - **Deep Learning Libraries**: - **PyTorch**: Popular deep learning framework. - **TensorFlow**: Leading deep learning framework. - **Keras**: High-level neural networks API. - **JAX**: Accelerated deep learning framework. - **Others**: Various other deep learning libraries. - **Automated Machine Learning**: - **AutoKeras**: Automated machine learning library. - **AutoML**: Automated machine learning tools. - **Natural Language Processing**: - **NLTK**: Natural Language Toolkit. - **spaCy**: Natural Language Processing library. - **Transformers**:

Overview: This project template provides a well-structured and best-practice approach for TensorFlow projects, focusing on simplicity, clear folder organization, and good Object-Oriented Programming (OOP) design. Key features: - Simple and well-organized project structure - Clear folder structure for easy navigation - Good OOP design principles - Includes a template file and simple example for beginners Tech stack: Python Why it matters: The template helps developers quickly get into their main project, focusing on the core components (models, training, etc.) without getting bogged down in unnecessary details. Repository: mgsalem/Tensorflow-Project-Template Primary language: Python GitHub stars: 3616 GitHub description: A best practice for TensorFlow project template architecture.

### Overview: Alink is a versatile machine learning algorithm platform built on Flink, developed by the PAI team at Alibaba Computing Platform. It offers a range of tools and resources for developers to build, deploy, and manage machine learning models efficiently. ### Key Features: - **Unified API**: Provides a single interface for various machine learning algorithms. - **Scalable Architecture**: Supports distributed processing across multiple nodes. - **Extensive Library**: Includes a wide range of algorithms for different tasks, such as classification, regression, clustering, and more. - **Integration with Flink**: Seamlessly integrates with Apache Flink for real-time and batch processing. ### Tech Stack: Java ### Why it Matters: Alink simplifies the process of developing, deploying, and managing machine learning models, making it easier for developers to build and scale their applications. Its unified API and integration with Flink ensure that it can handle complex tasks efficiently, making it a valuable tool in the development of data-driven applications.

### Overview: torchchat is a project showcasing the ability to run large language models (LLMs) using Python. It allows seamless integration of LLMs into applications, desktops, and mobile devices. ### Key features: - **Chat**: Direct interaction with LLMs for text-based conversations. - **Generate**: Text generation capabilities for various tasks. - **Browser**: Run chat in the browser for quick access. - **Desktop/Server Execution**: Run models without Python using AOT Inductor or the C runner. - **Mobile Execution**: Deploy and run on iOS and Android. ### Tech stack: Python ### Why it matters: torchchat provides a practical way to integrate large language models into various platforms, making them accessible and useful for developers and users alike.