Projects

### Overview: Wunjo is a powerful open-source tool for face swapping, lip syncing, object control, text removal, background restoration, and more. It's designed to be both local and free, making it accessible to everyone. ### Key features: - Face swapping - Lip syncing - Object control - Text removal - Background restoration - Restyling - Audio separator - Clone voice - Video generation ### Tech stack: C++ with Qt 6 and KF6 ### Why it matters: Wunjo offers a comprehensive set of features for video manipulation, making it a valuable tool for content creators, educators, and anyone looking to enhance their videos. Its open-source nature ensures it remains free and accessible to all, making it a versatile solution for various creative projects.

### Overview: HyperGAN is a composable GAN framework that simplifies the process of training, inspecting, and recovering GAN models. It supports native CUDA training, flexible Python components, and a functional API for easy integration into projects. ### Key Features: - **Composable GAN Framework**: Configurable and practical GANs with a user-friendly interface. - **Native CUDA Training**: Supports CUDA training for efficient performance. - **Flexible Python Components**: Easy integration with Python code. - **Complete Training Checkpoints**: Saves and loads training checkpoints for easy recovery. - **CLI for Live and Historical Runs**: Docker-style CLI for managing runs from any directory. - **Web Observatory**: Provides a local Web Observatory for monitoring and inspecting models. ### Tech Stack: - Python 3.12+ - PyTorch 2.14.0 ### Why It Matters: HyperGAN simplifies the process of training, inspecting, and recovering GAN models, making it easier for developers to integrate GANs into their projects. The framework supports native CUDA training, flexible Python components, and a functional API, making it a powerful tool for researchers and developers alike.

### Overview: This repository contains a curated collection of books and references for various fields of programming, including Computer Vision, Machine Learning, Deep Learning, NLP, Python, and more. It offers over 200+ programming ebooks at no cost, making it an excellent resource for anyone looking to enhance their knowledge in these areas. ### Key Features: - Comprehensive coverage of Computer Vision, Machine Learning, Deep Learning, NLP, Python, and more. - Over 200+ programming ebooks available in one place. - No cost for accessing the ebooks. - Includes detailed guides for beginners, such as a roadmap for AI/ML (AIML) and a practical AI/ML roadmap. ### Tech Stack: None specified. ### Why It Matters: This repository serves as a valuable resource for developers looking to expand their knowledge in various programming fields. The ebooks cover a wide range of topics and are available for free, making it an ideal choice for those seeking to improve their skills without the financial burden of purchasing expensive textbooks.

### Overview: This series will guide developers through the foundational concepts of Natural Language Processing (NLP) and Computer Vision, culminating in the cutting-edge Vision-Language Models (VLMs). The series will cover essential topics such as distributed word representations, sequence-to-sequence learning, attention mechanisms, and the development of early and modern vision-language models. ### Key Features: - **Word2Vec with Sub-Sampling and Negative Sampling in PyTorch** - **Seq2Seq Models** - **Attention Mechanisms** - **BERT and GPT** - **Early Vision-Language Models** - **CLIP** - **Modern Vision-Language Models** ### Tech Stack: Jupyter Notebook ### Why It Matters: This series will provide a comprehensive and practical guide for developers looking to understand and implement state-of-the-art Vision-Language Models. By following the tutorials and reading the accompanying papers, developers will gain the knowledge and skills necessary to tackle complex NLP and CV tasks, including image captioning, visual language understanding, and more.

### Overview: This repository offers a comprehensive set of simulated manipulation policy evaluation environments for real robot setups, specifically designed to complement real-world evaluations. It provides valuable quantitative metrics and standardized setups for evaluating robot manipulation policies. ### Key features: - **Real-to-Sim Evaluation**: The repository includes two real-to-sim evaluation setups, including visual matching and variant aggregation, to simulate real-world environments efficiently. - **SAPIEN and ManiSkill2 Benchmark Integration**: Utilizes the SAPIEN simulator and the ManiSkill2 benchmark, with GPU parallelization for faster execution. - **Standardized Environments**: Offers a standardized setup for reproducibility, making it easier to evaluate and compare different policy approaches. ### Tech stack: Python with Jupyter Notebook and PyTorch. ### Why it matters: This repository simplifies the evaluation of robot manipulation policies by providing efficient and scalable simulation environments, enabling researchers to test and compare their policies more effectively and efficiently.

### Overview: 이 리소스는 개발자들이 필요한 모든 정보를 한 곳에汇聚합니다. AI 도구, 공식 문서, 커뮤니티 등 다양한 리소스를 제공하여 개발 과정에서 필요한 모든 정보를 한 곳에서 찾을 수 있습니다. ### Key features: - [AI 개발 도구](#ai-dev-tools) - Cursor, Claude Code, bolt.new 등 - [AI 디자인 도구](#ai-design-tools) - Figma AI, Galileo 등 - [대화형 AI](#conversational-ai) - ChatGPT, Claude, Gemini, DeepSea - [개발 도구](#development-infra) - Git, Docker, Kubernetes 등 - [커뮤니티](#tech-community) - Stack Overflow, GitHub, Reddit 등 ### Tech stack: Node.js, React, Express, Docker, Kubernetes ### Why it matters: 이 리소스는 개발자들이 필요한 모든 정보를 한 곳에汇聚하고, 다양한 AI 도구와 커뮤니티를 제공하여 개발 과정에서 필요한 모든 정보를 한 곳에서 찾을 수 있습니다.

Overview: This repository contains datasets and code for the paper "Benchmarking Neural Network Robustness to Common Corruptions and Perturbations" by Dan Hendrycks and Thomas Dietterich, presented at ICLR 2019. Key features: - ImageNet-C datasets with 1000 classes and 224x224x3 images - Tiny ImageNet-C datasets with 200 classes and 64x64 images - CIFAR-10-C and CIFAR-100-C datasets for quicker experimentation - ImageNet-C leaderboard with robustness metrics for various methods Tech stack: Python 3+, PyTorch 0.3+ Why it matters: This repository provides a comprehensive benchmark for evaluating the robustness of neural networks to common corruptions and perturbations, helping researchers and practitioners understand the effectiveness of various robustness techniques.

### Overview: RAG Search API is a powerful tool for developers looking to integrate advanced search functionality into their applications. It provides a simple and efficient way to perform semantic search using the latest AI techniques. ### Key features: - **Semantic Search**: RAG Search API uses AI models to understand and interpret the context of search queries. - **Customizable Search Providers**: Supports multiple search providers, including Google, Bing, and more. - **Detailed Results**: Retrieves detailed search results with improved relevance and accuracy. - **Reranking**: Enhances the search results by re-ranking based on user preferences. ### Tech stack: Python ### Why it matters: RAG Search API is essential for developers who need a robust and efficient search solution that leverages the latest AI techniques to provide accurate and relevant results.

### Overview Punica is a system designed to serve multiple LoRA finetuned language models efficiently. It optimizes the storage and memory usage by leveraging the parameter efficiency of LoRA, which only adds a small percentage of storage and memory overhead. ### Key Features - **Efficient Storage and Memory Usage**: By using LoRA, Punica reduces the storage and memory footprint of each model by 1%. - **Batching Effect**: The pretrained model's batching effect is preserved, making it highly efficient. - **CUDA Kernel Optimization**: A CUDA kernel called Segmented Gather Matrix-Vector multiplication (SGMV) is used to efficiently compute the LoRA addon. - **High Throughput**: Punica achieves 12x throughput compared to state-of-the-art systems, making it highly scalable for multi-tenant environments. ### Tech Stack - Python - CUDA ### Why It Matters Punica is crucial for applications that require serving multiple LoRA finetuned language models, such as multi-tenant environments where each tenant needs a different model. Its ability to maintain the strong batching effect and high throughput makes it an essential tool for such scenarios.

### Overview: This repository contains practical AI prompts designed for product managers, leveraging tools like ChatGPT, Claude, and Gemini. The assets are ready-to-use templates that can be copied and pasted into these AI assistants to generate high-quality outputs in one go. ### Key Features: - **Product Requirement Document (PRD) Template**: Write a PRD from discovery notes. - **Jobs to Be Done Analysis**: Understand what customers actually need. - **Win/Loss Analysis**: Turn win/loss debriefs into ground truth. - **Stakeholder Map**: Map who matters and how to engage them. - **Incoming Request Breakdown**: Decode what an inbound ask really wants. - **Pre-mortem**: Kill the launch on paper before reality does. - **End-of-Life (EOL) Suite**: Write user stories engineers don't hate. - **Lean UX Canvas**: Frame an initiative as testable hypotheses. - **Agent Strategy Canvas**: Design an agentic AI system responsibly. - **Session Saver**: Save this session's context for the next one. ### Tech Stack: Python ### Why It Matters: These prompts equip product managers with the tools to create execution-quality outputs efficiently, ensuring they can focus on strategy without getting bogged down in the details.

Overview: AuditPilot is an enterprise AI agent platform designed specifically for audit workflows, offering a seamless integration of evidence-based auditing, controlled processes, tool execution, evaluation harnesses, human review, and remediation delivery. This platform aims to streamline the audit process by providing a traceable, reproducible, and explainable workflow. Key features: - Evidence-based auditing - Controlled processes - Tool execution - Evaluation harnesses - Human review - Remediation delivery Tech stack: Python Why it matters: AuditPilot simplifies the complex audit process by automating evidence collection, ensuring consistent and traceable results, and providing a structured workflow for human review and remediation, ultimately delivering a more efficient and reliable audit experience. Repository: Ricky-7-Yan/intelligent-audit-system GitHub stars: 1175 GitHub description: AuditPilot: auditable enterprise AI agents for evidence-grounded workflows, governed tools, evaluation harnesses, human review, and remediation delivery. This project is not just about saving time or improving accuracy; it's about delivering a robust, reproducible, and traceable audit process that meets the needs of modern enterprises.

### Overview: This repository is a curated list of foundation models for vision and language tasks. It provides a comprehensive overview of large-scale pretrained models that can be adapted to various applications, including vision and language tasks. ### Key features: - Foundation models for vision and language tasks - A curated list of pretrained models - Updated regularly to include the latest advancements in the field - Focuses on large-scale models for efficient adaptation ### Tech stack: Python, Markdown, and Git. ### Why it matters: Foundation models play a crucial role in accelerating research and development in artificial intelligence by providing pre-trained models that can be fine-tuned for specific applications. This repository serves as a valuable resource for researchers, developers, and practitioners looking to explore and leverage these models for their projects.

Overview: This repository is a comprehensive knowledge base for DevOps, security, infrastructure, and compliance skills, designed to empower AI agents with expert-level expertise in 160+ production-ready skills across various platforms and technologies. Key features: - 160+ production-ready skills across Kubernetes, Terraform, AWS/Azure/GCP, AI platform operations, container hardening, SOC2/ISO27001, and incident response. - Ready-to-run scripts, templates, and playbooks for SRE, platform, and security teams. - Skills are formatted in an Agent Skills format for easy integration with Claude Code, Cursor, Codex, and other supported agents. - Compatible with over 100+ agents, including Claude Code, Cursor, Codex, OpenCode, Cline, and many more. Tech stack: Shell Why it matters: This repository provides a valuable resource for developers and security professionals to enhance their skills, ensuring that their AI agents can handle complex infrastructure and security tasks effectively. GitHub stars: 1173 Repository: BagelHole/DevOps-Security-Agent-Skills

Overview: fastCRW is a fast, lightweight alternative to Firecrawl and Tavily, offering a lightweight Rust binary with a single binary size, 6 MB RAM, and self-host or managed cloud options. It provides a drop-in Firecrawl-compatible API for web scraping, crawling, and search functionalities, with a 2.3x faster performance compared to Tavily and 1.5x faster than Firecrawl in 1K-URL benchmarks. Key features: - Fast, lightweight web scraper, crawler, and search API - Self-host or managed cloud options - Drop-in Firecrawl-compatible API (/scrape, /crawl, /search) - 2.3x faster than Tavily, 1.5x faster than Firecrawl in 1K-URL benchmarks - 6 MB RAM, single binary - Rust-based, no credit card required - Managed proxies, JS rendering, and search capabilities - No credit card required - Managed proxies, JS rendering, and search capabilities - Runs locally as a small Rust binary or uses the managed API - 1000 free credits, no credit card needed - macOS, Linux, Intel, and ARM support - Automatically picks up Claude Code, Cursor, Codex, Gemini CLI, OpenCode, and Windsurf when already set up - No need to run or keep up with anything, key stays in `~/.config/crw/config.toml` - 90.0% answer accuracy on the Artificial Analysis Search Index, outperforming other products

### Overview: LLaMA-Mesh is a project that leverages large language models (LLMs) to generate 3D meshes, bridging the gap between text and 3D representation. ### Key Features: - **Unified Representation**: Represents vertex coordinates and face definitions of 3D meshes as plain text. - **Supervised Fine-Tuning**: Trains LLMs to generate 3D meshes from text prompts, understand and interpret 3D meshes, and produce interleaved text and 3D mesh outputs. - **Text-3D Integration**: Enables seamless integration of text and 3D data, leveraging spatial knowledge from LLMs. ### Tech Stack: - Python ### Why It Matters: LLaMA-Mesh is the first to demonstrate that LLMs can be fine-tuned to acquire complex spatial knowledge for 3D mesh generation in a text-based format, effectively unifying the 3D and text modalities. This approach offers key advantages of leveraging spatial knowledge embedded in LLMs and enabling conversational 3D generation and mesh understanding.

### Overview: The **clean-fid** repository provides a comprehensive and reproducible framework for evaluating generative models using the FID (Fréchet Inception Distance) metric. It addresses common issues and inconsistencies in FID calculations by implementing proper low-level image resizing and quantization steps. ### Key Features: - **Proper Image Resizing and Quantization**: Ensures consistent FID scores across different methods, papers, and groups. - **Cross-Implementation Support**: Supports multiple libraries and implementations, including TensorFlow, PyTorch, and OpenCV, ensuring compatibility and reproducibility. - **CLIP Feature Support**: Enables the calculation of FID using CLIP features, providing an additional layer of evaluation. - **Google Colab Notebooks**: Provides practical examples and demonstrations for computing FID and KID scores, including resizing and quantization steps. ### Tech Stack: - Python ### Why It Matters: The FID calculation involves many steps that can produce inconsistencies in the final metric. By providing a standardized and reproducible framework, **clean-fid** ensures that the FID scores are comparable across different methods, papers, and groups, making it a valuable tool for evaluating the quality of generative models. ### Repository: - **Repository**: GaParmar/clean-fid - **GitHub Stars**: 1171 - **Primary Language**: Python - **GitHub Description**: PyTorch - FID calculation with proper image resizing and quantization steps [CVPR 2022] ### Quick Start: - **Calculate FID**: [**

Overview: Neuro SAN Studio is a developer-friendly playground designed for building intelligent multi-agent systems. Key features: - Ready-to-run examples - Tutorials - Tools for designing, testing, and deploying agent networks - No installation required Tech stack: Python Why it matters: This studio simplifies the complex orchestration of neuro-synaptic networks, enabling researchers, developers, and domain experts to focus on problem-solving quickly.

Overview: This repository contains 80+ production-ready AI agent examples written in Python, covering autonomous AI agents, multi-agent systems, and agentic AI workflows using various tools such as uAgents, ASI:One, MCP, A2A, LangChain, CrewAI, Gemini, Claude, and OpenAI. Key features: - uAgents - ASI:One - MCP - A2A - LangChain - CrewAI - Gemini - Claude - OpenAI Tech stack: Python Why it matters: These examples provide a comprehensive set of production-ready AI agent projects that developers can use as starting points for building autonomous AI agents, multi-agent systems, and agentic AI workflows. GitHub stars: 1168 GitHub description: 80+ production-ready AI agent examples in Python — build autonomous AI agents, multi-agent systems and agentic AI with uAgents, ASI:One, MCP, A2A, LangChain, CrewAI, Gemini, Claude and OpenAI. Repository: fetchai/innovation-lab-examples Primary language: Python GitHub stars: 1168 GitHub description: 80+ production-ready AI agent examples in Python — build autonomous AI agents, multi-agent systems and agentic AI with uAgents, ASI:One, MCP, A2A, LangChain, CrewAI, Gemini, Claude and OpenAI.

### Overview: RTDL is a comprehensive collection of papers and packages focused on deep learning for tabular data. It includes benchmarking optimizers, analyzing data uncertainty, finetuning models, ensembling methods, nearest neighbor integration, and more. ### Key features: - **Benchmarking Optimizers for MLPs in Tabular Deep Learning**: Papers and code for evaluating different optimization techniques. - **Unveiling the Role of Data Uncertainty in Tabular Deep Learning**: Research on understanding the impact of data uncertainty. - **On Finetuning Tabular Foundation Models**: Papers and code for fine-tuning models on tabular data. - **TabM: Advancing Tabular Deep Learning with Parameter-Efficient Ensembling**: Papers and code for parameter-efficient ensembling methods. - **TabReD: Analyzing Pitfalls and Filling the Gaps in Tabular Deep Learning Benchmarks**: Papers and code for analyzing and improving tabular deep learning benchmarks. - **TabR: Tabular Deep Learning Meets Nearest Neighbors**: Papers and code for integrating tabular deep learning with nearest neighbor methods. - **TabDDPM: Modelling Tabular Data with Diffusion Models**: Papers and code for using diffusion models for tabular data. - **Revisiting Pretraining Objectives for Tabular Deep Learning**: Papers and code for revisiting and improving pretraining objectives. - **On Embeddings for Numerical Features in Tabular Deep Learning**: Papers and code for understanding and improving embeddings for numerical features. - **rtdl-revisiting-models**: A package for revis

### Overview: Cat Paper Collection is an academic paper repository that focuses on computer graphics, computer vision, and machine learning related to cats. It aims to provide a comprehensive collection of papers that produce experimental results related to cats, making it valuable for researchers and enthusiasts alike. ### Key features: - **Computer Graphics and Vision**: Papers that explore the visual aspects of cats, including their representation, analysis, and synthesis. - **Machine Learning**: Research that utilizes machine learning techniques to understand, model, and generate cat-related data. - **Collaboration**: The repository encourages contributions from authors and welcomes feedback to improve the collection. ### Tech stack: The repository uses Python for its data management and documentation. ### Why it matters: Understanding and synthesizing cats has become increasingly important in the field of computer vision and graphics. This collection provides a valuable resource for researchers and enthusiasts looking to advance in these areas, making it a significant contribution to the academic community.