Projects

### Overview: HIPFire is a Rust-based RDNA-native LLM inference engine designed for AMD GPUs. It offers a curated registry of models, a CLI for easy use, and a fast, efficient inference engine. ### Key features: - **Model Registry**: A curated registry of models optimized for AMD GPUs. - **CLI**: Command-line interface for quick access to models. - **GPU Support**: Native support for multiple GPUs, including RDNA and CDNA. - **Redline Replay Path**: A fast, efficient inference path for improved performance. - **Quantization**: Support for various quantization formats (MQ3–MQ6, MQ4R, MFP4). ### Tech stack: Rust + HIP + Redline. ### Why it matters: HIPFire provides a fast, efficient, and optimized inference engine for AMD GPUs, making it suitable for applications requiring high performance and low latency. Its curated model registry and fast inference path make it a valuable tool for developers looking to leverage the power of AMD GPUs in their projects.

### Overview: Build your own X - Master machine learning by creating everything from scratch. This project aims to cover a wide range of machine learning topics, including linear regression, logistic regression, neural networks, recommendation systems, computer vision, natural language processing, time series forecasting, anomaly detection, sentiment analysis, and more. ### Key Features: - Implementing core machine learning algorithms from scratch. - Comprehensive coverage of various machine learning techniques. - Projects include linear regression, logistic regression, K-Nearest Neighbors, Naive Bayes, Decision Trees, Random Forest, Support Vector Machines, K-Means Clustering, Principal Component Analysis, Perceptron, and Gradient Boosting. ### Tech Stack: Python ### Why it Matters: This project is designed to help developers understand and implement machine learning concepts from the ground up, equipping them with the skills necessary to build and deploy their own machine learning models.

Overview: Spectra is a Spec-Driven Development tool that combines a desktop app, a CLI, and skills for coding agents. It simplifies the workflow of managing specs, changes, and tasks, making it easier to track progress and automate the process. Key features: - View and edit specs and change proposals - Automatically parse task progress from Markdown checklists - Archive completed changes with one click - Intelligent scanning for contradictions and gaps across proposals, specs, designs, and tasks - Skills for coding agents: Claude Code, Codex, Cursor, GitHub Copilot, Antigravity, and JetBrains Junie Tech stack: TypeScript Why it matters: Spectra streamlines the process of managing specs and changes, making it easier to track progress and automate the workflow, ultimately improving productivity and reducing manual errors.

Overview: TryPost is an open-source social media scheduler designed for AI agents and teams. It allows users to create, schedule, and publish posts across multiple networks using either the MCP or REST API, or manually through a Composer. Key features: - **MCP and REST API**: Connects to AI assistants like Claude, ChatGPT, or Cursor to automate posts and analytics. - **Composer**: Writes posts once, adapts for different networks, and integrates with an AI assistant for rephrasing and approval. - **Calendar and Queues**: Manages posting times and reschedules posts, with a live preview and approval workflow. - **Approvals**: Teams can approve posts before they go live, with admin controls for brand-specific settings. Tech stack: PHP Why it matters: TryPost simplifies the process of creating and scheduling social media posts, making it accessible for both AI-driven and human-driven teams, while providing robust tools for automation and approval workflows.

Overview: DeepZero is an automated vulnerability research framework designed to find zero-day vulnerabilities by parsing, decompiling, and analyzing thousands of Windows kernel drivers using AI agents. Key features: - Pipeline-as-YAML: Defines pipelines declaratively. - Parallel execution: Uses ThreadPoolExecutor for parallelism. - Resumable runs: Atomic per-sample state on disk, supports re-running from saved state. - LLM integration: Supports Jinja2 prompt templates with any LLM provider via LiteLLM. - REST API (WIP): Experimental and incomplete. - Extensible: Write custom processors as Python classes. Tech stack: Python 3.11+. Why it matters: DeepZero automates the process of identifying exploitable IOCTLs in Windows kernel drivers, making it an invaluable tool for security researchers and developers looking to discover vulnerabilities in outdated software. Repository: 416rehman/DeepZero GitHub stars: 749 GitHub description: Find zero-days while you sleep. DeepZero is an automated vulnerability research framework that parses, decompiles, and analyzes thousands of Windows kernel drivers for exploitable IOCTLs natively using AI agents.

### Overview: An old coder's strategy for the agent era: don't read the code — make it run the gauntlet. Evidence-first development skill for coding agents, inspired by Uncle Bob. ### Key features: - **SPEC → GAUNTLET → EVIDENCE**: Concrete examples of what the code must and must not do, plus which tools the agent wants to install. Approving it is the single yes/no you give. - **EVIDENCE**: Real numbers from one final fresh run, rerunnable yourself with a single command. - **Full test suite**: Did anything break? - **Types + lint + complexity**: Any obvious mistakes? Any unreadable tangles? - **Changed-line coverage**: Is every new line actually exercised by a test? - **Mutation testing**: Plant bugs on purpose — do the tests catch them? - **Property-based tests**: Do the rules survive hundreds of random inputs? - **Real execution**: Does it actually run, outside the test harness? - **Supply chain & secrets**: Did the agent quietly pull in risky packages, or leak a key? - **Suite health**: Are the tests themselves stable, in any order? ### Tech stack: Markdown ### Why it matters: From Uncle Bob (Robert C. Martin), on working with coding agents: > My current strategy is to not read any of the code written by my agents. That’s the only way I can take advantage of their productivity. What I do instead is to surround the agents with extreme constraints. Unit tests, gherkin tests, QA procedures, quality metrics, mutation testing

### Overview: PMRF is a novel photo-realistic image restoration algorithm that provably approximates the optimal estimator that minimizes the Mean Squared Error (MSE) under a perfect perceptual quality constraint. This approach aims to achieve the lowest possible Mean Squared Error (MSE) while maintaining high perceptual quality, making it highly effective for image restoration tasks. ### Key features: - **Posterior-Mean Rectified Flow (PMRF)**: A state-of-the-art image restoration method that leverages the posterior mean to enhance image quality. - **Optimal Estimator**: The algorithm optimizes the MSE under a perceptual quality constraint, ensuring minimal error while maintaining high fidelity. - **Minimal MSE**: PMRF is designed to achieve the lowest possible MSE, making it highly effective for photo-realistic image restoration tasks. ### Tech stack: - Python ### Why it matters: PMRF is a groundbreaking algorithm that significantly reduces the Mean Squared Error (MSE) in image restoration tasks, making it highly effective for applications requiring high perceptual quality. Its ability to achieve minimal MSE while maintaining high fidelity makes it a valuable tool in the field of image processing and restoration.

Overview: Allosaurus is a pretrained universal phone recognizer capable of recognizing phones in over 2000 languages. Key features: - Recognizes phones in over 2000 languages - Supports both single audio file and directory inputs - Outputs both file names and phone sequences - Supports language-specific phone inventories for improved accuracy Tech stack: Python Why it matters: Allosaurus simplifies the process of recognizing phones in multiple languages, making it an essential tool for language researchers, linguists, and language enthusiasts. Repository: xinjli/allosaurus GitHub stars: 753

### Overview: The HELP WANTED: AI AGENTS project is a public notice board for gofrantic.com, where real bounties and money are posted for commercial work. The project aims to measure the quality of AI agents' work by verifying every claim, delivery, and payout against a public record. The town uses real bounties, money, and deadlines to test the capabilities of AI agents in a real-world commercial setting. ### Key Features: - Real bounties and money for commercial work - Verified claims, deliveries, and payouts - Public record of every claim, delivery, and payout - Human-driven, human-assisted, and fully autonomous agents welcome - Runx receipts bind machine-executed steps to receipts - Public thesis on acceptance rates, survival curves, and agent performance - Verified source records and receipts for claims where available ### Tech Stack: Shell ### Why It Matters: The project aims to measure the quality of AI agents' work by verifying every claim, delivery, and payout against a public record. This experiment challenges the industry's assumption that AI agents cannot perform real commercial work and provides insights into the capabilities and limitations of AI agents in a real-world setting.

### Overview: LightCompress is a powerful toolkit designed for compressing large models, including Language Models (LLMs), Vision-Language Models (VLMs), and Video Generative Models. It leverages advanced compression algorithms to enhance efficiency and reduce model size without compromising performance. ### Key Features: - **Model Compression**: Efficiently compresses models to save storage and improve inference speed. - **State-of-the-Art Algorithms**: Utilizes cutting-edge techniques for token reduction and quantization. - **Versatile Support**: Supports various models and tasks, including LLMs, VLMs, and video generative models. ### Tech Stack: - Python 3.11 ### Why It Matters: LightCompress simplifies the compression process, making it accessible for developers and researchers. It enhances model efficiency and reduces computational costs, enabling better performance on limited resources.

Overview: Agents Universe 是一个开源的企业级 AI Agent 平台，旨在解决智能体开发和协作的两极困局，通过多 LLM 统一接入、10 个开箱即用的智能体、Docker 一键自托管等功能，让智能体像人一样学习和工作。 Key features: - 多 LLM 统一接入 - 10 个开箱即用的智能体 - Docker 一键自托管 Tech stack: Python Why it matters: 通过让智能体像人一样学习和工作，Agents Universe 提供了一种新的方式来解决智能体开发和协作的两极困局，从而实现技术平权。 Repository: agents-universe/agents-universe Primary language: Python GitHub stars: 718 GitHub description: 不是问答机器人，而是能真正干活的数字分身。共享智能体，共享项目上下文，让项目所有成员一起协同工作。同时智能体会像人一样通过资料或者工作抽象和总结经验到项目上下文中。

### Overview: Scout is an open-source company intelligence agent designed to gather and organize knowledge from various sources, including Slack, Drive, and more. ### Key Features: - **Navigation over Search**: Scout navigates through different information sources like Slack, Drive, and more, rather than ingesting everything into a vector database. - **Self-Maintaining Wiki and CRM**: Scout maintains its own wiki and CRM, perfecting the information it learns from various sources. - **Persistent Memory**: Scout uses persistent memory to ensure context is always up-to-date and relevant. - **Context Providers**: Scout uses context providers to interact with different information sources, such as querying web pages, updating CRM, and more. ### Tech Stack: Python ### Why it Matters: Scout is designed to be a versatile and efficient agent for companies looking to improve their internal knowledge management and operational efficiency. By combining the power of natural language tools with a robust context provider system, Scout aims to streamline information gathering and organization, making it easier for teams to access and utilize the knowledge they need. ### Repository: agno-agi/scout ### GitHub Stars: 751 ### GitHub Description: Open Source Company Brain

Overview: This educational project provides a step-by-step guide to building a production-ready RAG application using FastAPI, Docker, and MongoDB. The project includes detailed explanations of various components and features, making it ideal for developers looking to learn and implement RAG applications. Key features: - FastAPI for backend development - Docker for containerization - MongoDB for data storage - Vector DB Factory for semantic search - Semantic Search for enhanced query capabilities - Augmented Answers for improved response quality Tech stack: Python Why it matters: This project is valuable for developers interested in creating robust and scalable RAG applications. By following the detailed steps provided in the videos and code, readers can learn how to build a production-ready RAG application using modern technologies like FastAPI, Docker, and MongoDB, ensuring their projects are both efficient and effective. GitHub stars: 750 Repository: bakrianoo/mini-rag

### Overview: This project is a mobile device-based face recognition solution that achieves real-time 2K detection with just 3.40M size and 8.5ms CPU time. It includes identification, detection, landmark, pose, alignment, attribute, and tracking capabilities. The solution is optimized for edge devices, making it highly practical for various applications. ### Key Features: - **Identification**: Achieves 99.653% accuracy with 3.41M size and 9ms CPU time. - **Detection**: Provides 20ms/50fps detection with 30M size. - **Landmark**: Produces 1ms CPU time with 5.7M size. - **Pose**: Offers <0.1ms CPU time with <1K size. - **Align**: Provides <0.1ms CPU time with <1K size. - **Attribute**: Achieves 14ms/71fps with 16.4M size. - **Tracking**: Real-time tracking with <2ms CPU time. ### Tech Stack: The project is built using MXNet, GluonCV, and DLib, with a total size of 3.40M. ### Why It Matters: This solution is particularly useful for applications requiring real-time face recognition on mobile devices, such as security systems, facial recognition, and surveillance. Its lightweight design and real-time performance make it ideal for edge devices with limited processing power. ### Repository: The repository is maintained by becauseofAI and has 75

### Overview: This repository compiles examples of high-profile failed machine learning projects, providing valuable insights into the potential pitfalls of AI. ### Key Features: - **Amazon AI Recruitment System**: Automated recruitment system canceled due to evidence of discrimination against female candidates. - **Genderify - Gender Identification Tool**: AI tool shut down after built-in biases and inaccuracies were discovered. - **Leakage and the Reproducibility Crisis in ML-based Science**: A team at Princeton University found significant errors in 329 ML-based science papers. - **COVID-19 Diagnosis and Triage Models**: Predictive models for diagnosing or triaging COVID-19 patients were found to be flawed. - **COMPAS Recidivism Algorithm**: Florida’s recidivism risk system found evidence of racial bias. - **Pennsylvania Child Welfare Screening Tool**: Predictive algorithm flagged a disproportionate number of Black children for 'mandatory' neglect investigations. - **Oregon Child Welfare Screening Tool**: Similar predictive algorithm for child welfare in Oregon was stopped after a month. - **U.S. Healthcare System Health Risk Prediction**: Widely used algorithm to predict healthcare needs exhibited racial bias. - **Apple Card Credit Card**: Credit card algorithm for Apple was investigated by financial regulators after customers complained of gender discrimination. ### Tech Stack: Unknown ### Why It Matters: These examples highlight the importance of thorough testing, fairness, and avoiding built-in biases in machine learning models. They serve as cautionary tales for developers and researchers to learn from and improve their models.

Overview: This repository is a curated list of public projects and practices built on Jev, TypeSafe AI's System One model for typed decisions. It answers two practical questions for developers: where Jev is making real decisions in live production workflows and which decision architectures can be cleanly copied and applied across different industries. Key features: - Production Validation: Where is Jev actively making real decisions in live production workflows? - Transferable Patterns: Which decision architectures can be cleanly copied and applied across different industries? Tech stack: Unknown Why it matters: This repository centralizes discussions about Jev, a model for typed decisions, and provides practical answers to developers' questions about its usage and transferable patterns across industries. Repository: v-modal/awesome-jev-tools GitHub stars: 748 GitHub description: A curated list of public projects and practices built on Jev, TypeSafe AI's System One model for typed decisions.

### Overview: PyWSD is a Python library that provides implementations of various Word Sense Disambiguation (WSD) technologies. It includes algorithms like Lesk, Maximizing Similarity, and baselines, along with information-content-based similarity metrics. The library is designed to be user-friendly and requires Python 3.10+. ### Key Features: - **Lesk Algorithms**: Includes original, adapted, and extended versions of the Lesk algorithm. - **Maximizing Similarity**: Supports path-based and information-content-based similarity metrics. - **Baselines**: Provides random sense, first WordNet sense, and highest lemma counts as baselines. - **Information-Content Files**: Bundled with the package, including frequency dictionaries for Resnik, Jiang-Conrath, and Lin similarity metrics. - **Synset Signatures**: Computes synset signatures on demand for efficient repeated calls. ### Tech Stack: - Python 3.10+ ### Why It Matters: PyWSD is a valuable tool for developers and researchers interested in improving the accuracy of word sense disambiguation techniques. It provides a comprehensive set of algorithms and metrics that can be used to enhance the performance of WSD systems. The library's simplicity and ease of use make it an excellent choice for both beginners and advanced users. ### Repository: - **GitHub**: [alvations/pywsd](https://github.com/alvations/pywsd) - **Stars**: 749 - **Description**: Python Implementations of Word Sense Disambiguation (WSD) Technologies

### Overview: InstructIR is a novel approach that leverages human-written instructions to enhance image restoration, achieving state-of-the-art results across various tasks including denoising, deraining, deblurring, dehazing, and low-light image enhancement. ### Key features: - 📝 Human-written instructions guide the image restoration process. - 🖥️ State-of-the-art results on multiple restoration tasks. - 📈 +1dB improvement over previous all-in-one restoration methods. - 📈 Novel benchmark dataset and results for text-guided image restoration and enhancement. ### Tech stack: Jupyter Notebook ### Why it matters: InstructIR offers a unique solution for image restoration by leveraging human feedback, making it highly effective and versatile for a wide range of image degradation tasks.

Overview: This repository contains a curated list of books, organized by genres and authors, providing a comprehensive guide for book enthusiasts. Key features: - Genre-specific book lists - Author-based recommendations - Easy navigation through tags and filters Tech stack: Markdown Why it matters: This repository serves as a valuable resource for anyone looking to explore different genres and authors, offering a structured and organized way to discover new books.

Overview: Mask3D is a state-of-the-art model designed for accurate 3D semantic instance segmentation, achieving top performance on ScanNet, ScanNet200, S3DIS, and STPLS3D datasets. Key features: - Accurate 3D semantic instance segmentation - Supports ScanNet, ScanNet200, S3DIS, and STPLS3D datasets - Utilizes the MinkowskiEngine framework for efficient computation - Preprocessed datasets available for easy setup Tech stack: Python Why it matters: Mask3D's modular architecture and efficient computation make it a powerful tool for researchers and developers working on 3D semantic instance segmentation tasks. Repository: JonasSchult/Mask3D Primary language: Python GitHub stars: 748 GitHub description: Mask3D predicts accurate 3D semantic instances achieving state-of-the-art on ScanNet, ScanNet200, S3DIS and STPLS3D.