Projects

Overview: This repository contains a collection of papers, datasets, conferences, workshops, and companies related to the field of fashion and AI. The papers are sorted by their arXiv submission date, and the repository is constantly updated to provide the latest resources in the field. Key features: - Papers: A curated list of papers related to fashion and AI, sorted by submission date. - Related Events: Information on conferences, workshops, and companies related to fashion and AI. - Datasets: A list of datasets used in fashion and AI research. - Companies: Companies involved in fashion and AI. - Other Useful Resources: Additional resources that might be helpful for fashion and AI enthusiasts. Tech stack: Shell Why it matters: This repository serves as a valuable resource for anyone interested in the intersection of fashion and artificial intelligence, providing access to cutting-edge research, datasets, and industry insights.

### Overview: This project is a PyTorch implementation of Fully-Convolutional Siamese Networks for Object Tracking, based on the original paper "Fully-Convolutional Siamese Networks for Object Tracking" by Lu et al. The project includes training, tracking, and visualization components, and is available on GitHub with 637 stars. ### Key features: - PyTorch implementation - Training and tracking functionalities - Baseline results close to the original authors' performance - Baseline model available as *BaselinePretrained.pth.tar* ### Tech stack: Python ### Why it matters: This project provides a practical and accessible way to implement and test the Siamese network tracking algorithm, which is crucial for real-time object tracking applications.

### Overview: KoSpeech is an open-source toolkit designed for end-to-end Korean automatic speech recognition (ASR) using PyTorch and Hydra. It provides a comprehensive framework for developing and evaluating speech recognition models, including various models such as Deep Speech 2, LAS, Joint CTC-Attention LAS, RNN-Transducer, Speech Transformer, and Jasper. ### Key Features: - **Model Support**: Supports multiple models including Deep Speech 2, LAS, Joint CTC-Attention LAS, RNN-Transducer, Speech Transformer, and Jasper. - **Modular Architecture**: Provides a modular and extensible design for easy integration and modification. - **Documentation**: Detailed documentation is available, including installation instructions, model descriptions, and usage examples. - **Preprocessing**: Includes preprocessing methods for the KsponSpeech corpus, making it suitable for Korean ASR research. ### Tech Stack: Python ### Why It Matters: KoSpeech serves as a valuable resource for researchers and developers interested in Korean ASR. It offers a platform to experiment with different models and provides a standardized way to evaluate and compare performance. The toolkit's modular design and extensive documentation make it accessible and useful for both beginners and advanced researchers. ### GitHub Stars: 637 ### Repository: sooftware/kospeech

### Overview: 本教程全面指导你快速搭建自己的AI应用环境，从Docker桌面版的安装与配置开始，到本地部署Dify并自定义AI助手功能，实现多种特色AI应用。教程涵盖基础智能体、工作流、知识库、DeepResearch、数据库、MCP、复杂任务编排等高阶任务，由浅入深地学习掌握基于Dify的大模型应用开发。 ### Key features: 1. **Docker本地部署**：快速启动AI应用服务。 2. **提示词设计**：学习如何给AI下达精准指令。 3. **哄哄模拟器**：设计情感交互的游戏。 4. **新生入学指南助手**：高效检索知识库，创建信息问答应用。 5. **快速批处理小问题**：批量任务编排，数据处理分析。 6. **小红书读书卡片**：创建和配置Agent工作流，优化任务执行效率。 7. **面试宝典**：创建高效知识库索引，实现专业咨询与检索。 8. **text2sql及echart数据分析**：学习数据与AI交互，打开新思路。 9. **DeepResearch设计**：快速上手DeepResearch应用项目。 10. **MCP配置与应用**：掌握MCP客户端配置与原理。 11. **多模态入门（语音文本转化）**：快速完成语音智能体开发。 ### Tech stack: 本教程使用Docker、Dify、提示词设计、工作流、知识库、

### Overview: Discover and explore a curated list of agent-native services and infrastructure designed for AI agents. This repository includes email, browsers, memory, sandboxes, payments, and MCP tools, providing a selection criteria and onboarding guides to help you connect your agent. ### Key Features: - **Communication Services**: Agent-native email and identity. - **Browser & Web Execution**: Remote browsers for agents. - **Tool Access & Integration**: MCP servers and agent tools. - **Oversight & Approval**: Human-in-the-loop for agents. - **Commerce & Payments**: Agent wallets and protocols. - **Agent Runtime & Infrastructure**: Sandboxes, gateways, and identity. - **Agent Harnesses & Operator Surfaces**: Multi-agent control and live Codex HUDs. - **Memory & State**: Long-term memory for LLM agents. - **Search & Web Intelligence**: LLM-ready retrieval. - **Code Execution**: Secure sandboxes for AI-generated code. - **Observability & Tracing**: Agent trajectories and evals. - **Durable Execution & Scheduling**: Long-running agent workflows. - **Meeting & Conversation**: Bots and voice rooms for agents. - **Voice & Phone**: Telephony for AI agents. - **LLM Gateway & Routing**: Cost control and model routing. - **Agent Social & Community**: MCP-native social layers. ### Tech Stack: Python ### Why It Matters: This curated list provides a comprehensive overview of agent-native services and infrastructure, tailored specifically for AI agents. It helps developers and AI agents find the right tools and services designed for their

Overview: klaw is an enterprise AI agent orchestration solution that provides Kubernetes-like management, monitoring, and scaling capabilities for AI agents. It simplifies the deployment, visibility, isolation, scheduling, and debugging of AI agents. Key features: - **Visibility**: Check the status of AI agents using `klaw get agents` or `klaw logs`. - **Isolation**: Agents run in isolated namespaces with scoped permissions. - **Scheduling**: Automate agent scheduling with `klaw cron create`. - **Scaling**: Automatically scale agents using `klaw node join`. - **Debugging**: Real-time logs with `klaw logs agent --follow`. Tech stack: Go Why it matters: klaw simplifies the management, monitoring, and scaling of AI agents, making it easier to manage complex AI environments and ensure high availability and performance. Repository: klawsh/klaw.sh Primary language: Go GitHub stars: 635 GitHub description: kubectl for AI Agents README excerpt: <p align="center"> <img src="docs/assets/klaw-logo.png" alt="klaw" width="140" /> </p> <h1 align="center">klaw</h1> <p align="center"> <strong>kubectl for AI Agents</strong> </p> <p align="center"> Enterprise AI agent orchestration. Manage, monitor, and scale your AI workforce.<br/> One binary. Deploys in seconds. Scales to hundreds of agents. </p> <p align="center">

Overview: The Roboflow Python package provides a comprehensive suite for managing and interacting with Roboflow datasets, models, and projects, enabling developers to build and deploy computer vision applications efficiently. Key features: - Upload and manage images, annotations, and datasets - Start training vision models on Roboflow - Run inference on models hosted on Roboflow or self-hosted via Roboflow Inference - Manage projects and datasets Tech stack: Python Why it matters: This package simplifies the process of integrating Roboflow into projects, making it easier for developers to leverage Roboflow's powerful tools and resources for computer vision applications. Repository: roboflow/roboflow-python GitHub stars: 636 GitHub description: The official Roboflow Python package. Manage your datasets, models, and deployments. Roboflow has everything you need to build a computer vision application.

Overview: SandBase Harness is a local-first, self-hosted AI agent runtime and MCP bridge that provides sandboxed sessions, memory, credentials, audit/replay, and a built-in Console. It includes SQLite metadata, resumable events, MCP toolsets, and a TypeScript SDK. Key features: - Claude Managed Agents-style `/v1` API and local Console - Persistent sessions, resumable events, audit, and replay - MCP toolsets, permission policies, built-in tools, and skill packages - Local-first SQLite and file storage with no required hosted control plane Tech stack: TypeScript Why it matters: SandBase Harness simplifies the development and deployment of AI agents by providing a robust and secure environment for running and inspecting long-running models.

### Overview: Termio is a terminal-first agentic development environment designed for agents to write most of the code, providing a real terminal experience. It runs on macOS, Linux, and supports remote access via SSH, making it a versatile development tool. ### Key features: - **Terminal-first Environment**: Runs on macOS, Linux, and supports remote access via SSH, ensuring a consistent terminal experience across different platforms. - **Agent Hooks**: Automatically wires up each agent's hooks on first launch, providing real-time feedback and notifications. - **Session Management**: Sessions report working, idle, or "needs you" and provide visual cues like sidebars, menus, and notifications. - **tmux Alternative**: Runs in a daemon process, detaching only when the app is closed, and supports remote sessions via SSH. - **Remote Access**: Supports any Linux VPS, allowing users to access their development environment from anywhere. - **Project Management**: Projects hold sessions, and worktrees nest under the project they came from, facilitating parallel task management. - **Split Panes**: Offers split panes for multiple agents, development servers, and shells in one window, enhancing productivity. ### Tech Stack: Swift ### Why it Matters: Termio is a powerful development environment designed for agents to write most of the code, providing a real terminal experience. Its features, such as session management, agent hooks, and remote access, make it a versatile tool for developers working on macOS, Linux, and remote environments.

**Overview:** CookHero 是一个基于 LLM、RAG、Agent、多模态和营养数据分析的智能饮食与烹饪管理平台，旨在帮助厨房新手成为“烹饪英雄”。它提供智能菜谱查询、个性化饮食计划、AI 饮食记录、营养分析、Web 搜索增强等功能，支持多轮对话理解和精准行动建议。 **Key features:** - **智能问答**：解答烹饪技巧、食材搭配、营养知识等问题 - **个性化推荐**：根据口味、目标和限制提供更贴合的菜品选择 - **饮食计划**：按周规划三餐与加餐，形成可执行的饮食节奏 - **AI 记录**：文字/图片一键记录，自动估算热量与宏量营养 - **营养分析**：每日/每周统计与计划偏差分析，持续优化习惯 - **深度理解**：多轮对话理解用户意图，输出精准行动建议 - **实时搜索**：结合 Web 搜索获取最新烹饪资讯和趋势 **Tech stack:** Python, FastAPI, LangChain, Milvus, NeMo Guardrails, RAGAS, PostgreSQL, Redis, MinIO **Why it matters:** CookHero 是一个集成了多种先进技术的智能饮食与烹饪管理平台，通过提供个性化饮食管理、智能菜谱查询、AI 饮食记录、营养分析等功能，帮助用户实现健康、智能的饮食习惯，成为厨房中的“烹饪英雄”。

### Overview: Avibe is a local-first Agent OS that allows users to run the official Claude, Codex, and OpenCode from their browser, chat app, or any other device. It simplifies the process of using these powerful AI models by making them accessible and controllable from a single terminal window. ### Key features: - **Runs the official Claude, Codex, and OpenCode**: Avibe integrates these models seamlessly into the Workbench, allowing users to interact with them directly from their browser or chat app. - **Built-in Vibey agent**: Avibe ships its own official agent, Vibey, which handles scheduling tasks, monitoring watches, and managing show pages. It integrates with Claude Code, Codex, and OpenCode, making it a versatile tool for AI developers. - **Simplified command execution**: Vibey condenses long commands into Watches, allowing users to interact with models without leaving their current session, making it easier to manage and monitor AI work. ### Tech stack: Python ### Why it matters: Avibe simplifies the process of using powerful AI models like Claude, Codex, and OpenCode, making them accessible and controllable from a single terminal window. It integrates these models seamlessly into the Workbench, allowing users to interact with them directly from their browser or chat app. This approach makes it easier to manage and monitor AI work, and it simplifies the process of using these models, making it more user-friendly and efficient.

### Overview: BrowserPilot is an intelligent web browsing agent that understands natural language commands, making it easier and more intuitive to automate tasks on the web. It uses GPT-3 to translate human instructions into executable code, allowing for seamless and precise automation without the need for complex programming. ### Key Features: - **Natural Language Control**: Commands are given in plain English, making it easy to understand and use. - **Selenium Integration**: Works seamlessly with Selenium, ensuring compatibility with modern web applications. - **Prompt Library**: A library of pre-written prompts for common tasks, reducing the need for extensive coding. - **Prompt Writing**: Detailed instructions on how to write effective prompts, including examples and best practices. ### Tech Stack: Python ### Why It Matters: BrowserPilot simplifies the process of automating web tasks by leveraging natural language understanding, making it accessible to users without extensive technical knowledge. This tool enhances productivity and efficiency in web automation, making it a valuable addition to the developer community. ### Repository: handrew/browserpilot ### Primary Language: Python ### GitHub Stars: 635 ### GitHub Description: Natural language browser automation

### Overview: SINQ is a fast, model-agnostic quantization method designed to make Large Language Models smaller while preserving accuracy. It was accepted at ICML 2026 and is now integrated into HF Transformers, making it easy to use in Hugging Face models. ### Key Features: - **Fast Quantization**: SINQ can quantize models in just a few seconds, making it suitable for real-time applications. - **Model Agnostic**: SINQ can be applied to any LLM model, not just Qwen3-14B or DeepSeekV2.5-236B. - **State-of-the-Art Performance**: SINQ achieves state-of-the-art performance without sacrificing accuracy. - **Easy Integration**: SINQ is now integrated into HF Transformers, making it accessible for developers. ### Tech Stack: - **Python**: The primary language used for implementing SINQ. ### Why It Matters: - **Efficient Memory Usage**: SINQ allows deploying large models on GPUs that would otherwise be too large, reducing memory usage significantly. - **Performance Preservation**: SINQ preserves the accuracy of LLMs while making them smaller, making it a powerful tool for model deployment. ### Repository: huawei-csl/SINQ - **GitHub Stars**: 635 - **Primary Language**: Python - **GitHub Description**: Welcome to the official repository of SINQ! A novel, fast and high-quality quantization method designed to make any Large Language Model smaller while preserving accuracy [ICML 2026] ### How to Reproduce

Overview: Lemma is a multiplayer harness that allows teams to share and collaborate on projects through agents, apps, and workflows. It integrates seamlessly with popular communication tools like Slack, Telegram, WhatsApp, and Teams, and can be hosted on Claude Code, Codex, or any OpenAI-compatible key. Key features: - Seamless integration with popular communication tools - Ability to share and collaborate on projects through agents, apps, and workflows - Supports multiple models and OpenAI-compatible keys - Built-in skills for building teammates based on user's preferences Tech stack: Python Why it matters: Lemma simplifies project collaboration by providing a unified platform that leverages existing communication tools, making it easy for teams to share and work on projects together.

### Overview: This repository contains a comprehensive collection of papers related to NeRF (Neural Radiance Fields), focusing on papers presented at various computer vision conferences including CVPR, ICCV, ECCV, NIPS, ICML, and ICLR. ### Key Features: - Papers from CVPR, ICCV, ECCV, NIPS, ICML, and ICLR - Detailed summaries and analyses of NeRF-related research - Comprehensive coverage of NeRF advancements and applications ### Tech Stack: - Papers from various computer vision conferences ### Why it Matters: This repository serves as a valuable resource for researchers and developers interested in the latest advancements in NeRF technology, providing a clear overview of the field's development and applications across different conferences and years. ### Repository: Repository: lif314/awesome-NeRF-papers Stars: 635 Description: All NeRF-related papers @ CVPR/ICCV/ECCV/NIPS/ICML/ICLR

### Overview: TrustLLM is a Python-based toolkit designed to evaluate the trustworthiness of large language models across six key dimensions: truthfulness, safety, fairness, robustness, privacy, and ethics. It provides a comprehensive framework for benchmarking and automating evaluations using local models, model APIs, and AI agent orchestration. ### Key Features: - **Benchmarking**: Runs the ICML 2024 benchmark with local weights or model APIs. - **Data Management**: Keeps the data, settings, and results together. - **AI Agent Integration**: Guides CLI/Python orchestration and JSON results. - **Documentation**: Provides comprehensive documentation and a machine-readable documentation index. ### Tech Stack: - Python ### Why It Matters: TrustLLM offers a practical and reproducible method for evaluating the trustworthiness of large language models, ensuring that models are reliable, safe, fair, robust, private, and ethical. It streamlines the evaluation process, making it easier to assess the performance of different models across various dimensions. ### Repository: - **Repository**: HowieHwong/TrustLLM - **GitHub Stars**: 632 - **Primary Language**: Python - **GitHub Description**: An open research toolkit for evaluating the trustworthiness of large language models across six dimensions. Supports local models, model APIs, and AI agent orchestration. ### Primary Language: - **Primary Language**: Python ### GitHub Description: - TrustLLM is an open research toolkit for evaluating the trustworthiness of large language models across **six dimensions**. It

Overview: This project is an unofficial PyTorch reimplementation of RandAugment, a technique that enhances image classification models by randomly applying augmentations. It simplifies the process of training models with RandAugment without the need for a separate proxy task. Key features: - Unofficial PyTorch reimplementation of RandAugment - Includes Fast AutoAugment code for inspiration - Supports CIFAR-10, CIFAR-100, and SVHN datasets - Provides a simple usage example with a transform function - Reproduces similar results as the original paper Tech stack: Python Why it matters: RandAugment is a powerful technique for improving the robustness and performance of image classification models, making it an essential tool for researchers and practitioners in the field.

Overview: ai-business-skills is a comprehensive set of marketing skills for AI agents, designed to help businesses optimize their marketing efforts across multiple regions and languages. Key features: - 144 bilingual skills (VN + Global) - 4 role-specific SOP packs (content, design, performance, leader ops) - 6 agents covering 19 workflows - 4 regions (US/EU/SEA/LATAM) + Vietnam - 2025-2026 strategy and personal branding - 2025-2026 knowledge library - Companion: opa-kit Tech stack: PowerShell Why it matters: ai-business-skills provides a structured approach to marketing, ensuring that businesses can effectively communicate their products and services across diverse markets and languages, leveraging the strengths of AI agents and tailored workflows for optimal results. Repository: minhnv0807/ai-business-skills GitHub stars: 628 GitHub description: 138 bilingual AI marketing skills (69 VN + 69 Global) for Claude Code, OpenCode, Codex, VS Code. Four role SOP packs — content, design, performance, leader ops — plus strategy, personal brand, AI avatar, dropshipping, design master, knowledge library. 4 regions (US/EU/SEA/LATAM) + Vietnam 2025-2026. Companion: opa-kit. Repository: minhnv0807/ai-business-skills

### Overview: Waggle is a lightweight token-based handoff mechanism for multi-agent systems, designed to overcome the limitations of traditional handoffs and provide a more efficient and accurate way to share context between agents. ### Key features: - **Token-based Handoff**: Waggle uses a 30-byte token instead of pasting context to every subagent, ensuring accurate and efficient context sharing. - **Counts Every Read**: Every read of the token counts towards the token's budget, ensuring accurate consumption. - **Propagates Corrections**: The token propagates corrections to every holder, ensuring context is always up-to-date. - **No Per-Agent Shaping**: No need for per-agent shaping or receipts, making it easier to manage. - **No Server Needed**: No need for a server, as the token can be resolved locally. - **No Lifecycle Issues**: No stale paths, ensuring data is always fresh. - **No Telemetry Issues**: No issues with telemetry, as the token is resolved locally. ### Tech stack: Rust ### Why it matters: Waggle provides a more efficient and accurate way to share context between agents, overcoming the limitations of traditional handoffs and providing a more reliable and accurate way to share context between agents.

### Overview: The Kalshi AI Trading Bot is an agent-native toolkit designed for Kalshi prediction markets. It measures whether you actually beat the sharp Kalshi book, providing a rigorous way to prove or disprove your edge. ### Key Features: - **Edge Measurement**: Tracks and scores your trades against the sharp Kalshi book, providing a Brier score, log-loss, and edge vs. book metrics. - **Agent-Native Surface**: Offers a CLI, MCP server, and Claude skill for agent-native interface. - **SQLite Telemetry**: Tracks position and trades in SQLite. - **Streamlit Dashboard**: Provides a visual dashboard for tracking and analyzing your trades. - **OpenRouter Model Client**: Supports any model on OpenRouter for LLM clients. - **Atomic CLI Tools**: Includes a first-mover MCP server and a Claude skill for agent-native surface. - **Example Strategies**: Forkable example strategies for starting points. ### Tech Stack: - Python 3.12+ - Signed Kalshi API client - Market-data ingestion - Position tracking - SQLite telemetry - Streamlit dashboard - OpenRouter model client - Atomic CLI tools - Claude skill ### Why It Matters: This toolkit is designed to measure whether you have edge against the sharp Kalshi book, providing a rigorous way to prove or disprove your edge. It refutes the niche's central failure of asserting skill from a handful of lucky settlements, ensuring that the measurement is based on settled reality and out-of-sample honesty.