Projects

### Overview: pg-aiguide is an AI-optimized PostgreSQL expertise tool designed to help AI coding agents generate better Postgres code. It provides semantic search across official PostgreSQL manuals, curated Postgres best practices, and extension ecosystem docs. ### Key Features: - Semantic search across official PostgreSQL manuals (version-aware) - AI-optimized "skills" for curated Postgres best practices - Extension ecosystem docs, starting with TimescaleDB, with more coming soon ### Tech Stack: Python ### Why it Matters: pg-aiguide fixes outdated, missing constraints, and indexes, and provides modern Postgres patterns, making AI-generated Postgres code more robust, performant, and maintainable. ### Repository: timescale/pg-aiguide ### Primary Language: Python ### GitHub Stars: 1873 ### GitHub Description: MCP server and Claude plugin for Postgres skills and documentation. Helps AI coding tools generate better PostgreSQL code.

Overview: Automl-gs is an AutoML tool designed for citizen data scientists and engineers to create high-performing machine learning or deep learning models without needing to know advanced statistical techniques or machine learning engineering. It generates Python code for models and integrates seamlessly into prediction workflows, providing detailed metrics and easy model tweaking. Key features: - Generates native Python code - Handles messy datasets - Retrains models on new data without changes - Supports multiple ML/DL frameworks - Runs on Google Colab with TPUs Tech stack: Python 3, TensorFlow, XGBoost, scikit-learn Why it matters: Automl-gs simplifies the process of creating powerful prediction workflows without requiring deep knowledge of modern data preprocessing and machine learning engineering techniques. It optimizes time spent on model performance optimization, making it ideal for citizen data scientists and engineers. Repository: minimaxir/automl-gs Primary language: Python GitHub stars: 1868

Overview: This is a Keras port of the Single Shot MultiBox Detector (SSD) model architecture introduced by Wei Liu et al. in the paper "SSD: Single Shot MultiBox Detector". The repository provides accurate ported weights and models trained from scratch, making it easier to understand and adapt the model for various applications. Key features: - SSD300 - SSD512 - SSD7 (a smaller 7-layer version for faster training on mid-tier GPUs) Tech stack: Python Why it matters: This implementation provides detailed documentation and comments, making it easier to understand and adapt the model for various applications, especially for those interested in a low-level understanding of the model. Repository: pierluigiferrari/ssd_keras GitHub stars: 1867 GitHub description: A Keras port of Single Shot MultiBox Detector

### Overview: Liquid Time-Constant Networks (LTCs) are continuous-time models designed for backpropagation through time (BPTT). This repository provides a comprehensive solution for training and evaluating continuous-time models using TensorFlow 1.14.0 and Python 3 on Ubuntu 16.04 and 18.04 machines. The repository includes models for various applications such as hand gesture segmentation, room occupancy detection, human activity recognition, traffic volume prediction, and ozone level forecasting. Each model is trained and evaluated using specific scripts that accept parameters such as the model type, number of epochs, and hidden units. ### Key Features: - **Continuous-Time Models**: Implementations of Liquid Time-Constant Networks, Neural Ordinary Differential Equations (ODEs), Continuous-time Recurrent Neural Networks (CRNNs), and Continuous-time Gated Recurrent Units (GRUs). - **Training Scripts**: Scripts for training continuous-time models with backpropagation through time (BPTT) for various datasets including hand gesture segmentation, room occupancy detection, human activity recognition, traffic volume prediction, and ozone level forecasting. - **Model Parameters**: Optimized parameters for each model type, including minibatch size, learning rate, hidden units, optimizer, beta_1 and beta_2, number of epochs, BPTT length, and ODE solver settings. - **Results Storage**: Results are stored in the `results` folder, including best-performing checkpoints and evaluation metrics for each model. ### Tech Stack: - Python 3 - TensorFlow 1.14.0

### Overview: AgentsMeetRL is an extensive list of open-source repositories dedicated to training language model agents using reinforcement learning. This repository provides a comprehensive overview of various projects that focus on multi-turn interactions, tool use, and other advanced RL techniques. ### Key Features: - **Base Framework**: General-purpose RL training frameworks. - **General/MultiTask**: Agent systems trained across multiple tasks or environments. - **Search & RAG**: Search-augmented reasoning agents using retrieval tools. - **Web & GUI**: Agents interacting with web browsers, mobile/desktop GUIs, or operating systems. - **Tool-Use**: Agents trained to invoke external tools. - **Code & SWE**: Software engineering and code generation agents. - **Reasoning**: Reasoning agents with multi-turn or tool-integrated reasoning. - **Multi-Agent RL**: Multi-agent collaboration, negotiation, or credit assignment via RL. - **Memory**: Agents managing, retrieving, or evolving memory. - **Embodied**: Agents operating in embodied/physical simulation environments. - **Domain-Specific**: RL agents for specialized domains. - **Reward & Training**: Process/outcome reward models and training methodologies. ### Tech Stack: HTML ### Why It Matters: This repository serves as an invaluable resource for developers and researchers interested in the latest advancements in language model agent training using reinforcement learning. It provides a detailed overview of various projects, helping users understand the technical choices and methodologies used in these projects.

### Overview: This curated collection of research papers covers core topics in the AI agent ecosystem, including multi-agent coordination, memory & RAG, tooling, evaluation & observability, and security. It provides insights into the latest advancements and helps stay updated with the field's direction. ### Key Features: - **Multi-Agent Coordination**: Papers on how to effectively coordinate multiple agents. - **Memory & RAG**: Research on how agents use memory and Retrieval-Augmented Generation (RAG) techniques. - **Tooling**: Papers on tools and frameworks for AI agent development and deployment. - **Evaluation & Observability**: Insights into evaluating agent performance and improving observability. - **Security**: Papers on securing AI agents and ensuring robust security measures. ### Tech Stack: Unknown ### Why It Matters: This list helps AI engineers, researchers, and developers stay informed about the latest advancements in AI agent technology, ensuring they are up-to-date with the field's direction and best practices.

Overview: 本项目是一款基于大模型的自动化代码审查工具，旨在提升代码质量和开发效率。 Key features: - 多模型支持 - 消息即时推送 - 自动化日报生成 - 可视化 Dashboard - 选择性 Review Style - Agentic Review 模式（可选） Tech stack: Python Why it matters: 通过引入大模型，该工具能够提供更准确和快速的代码审查，提高团队的协作效率和代码质量。

### Overview: CoreCoder is a minimal AI coding agent written in Python, inspired by Claude Code. It consists of 1,322 lines of code, making it a living reference for understanding how coding agents function. The engine is designed to be easily readable and hackable, with breakpoints, changes, and reruns possible on any line. ### Key Features: - **Readable End to End:** The entire engine can be read in an afternoon, with no magic hidden. - **Hackable:** Set breakpoints, change lines, and rerun the code on your own machine. - **Minimal Core:** The engine is intentionally kept minimal, focusing on the core functionality without unnecessary features. ### Tech Stack: Python ### Why It Matters: CoreCoder provides a foundational understanding of how coding agents work, allowing developers to learn and improve upon it. It serves as a reference for how agents operate, making it easier to create and improve upon similar tools.

### Overview: WER are we? is an attempt at tracking the state of the art and recent results in speech recognition. It aims to provide an up-to-date overview of the field by comparing various models and techniques. ### Key Features: - **LibriSpeech**: Trained on more data than LibriSpeech. - **WER Test-Clean**: 5.83% (Deep Speech 2: End-to-End Speech Recognition in English and Mandarin) - **WER Test-Other**: 12.69% (HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units) - **Paper**: Humans [Deep Speech 2: End-to-End Speech Recognition in English and Mandarin] - **Published**: December 2015 - **Notes**: *Humans* ### Tech Stack: - **LibriSpeech**: Trained on more data than LibriSpeech. ### Why It Matters: WER are we? is an attempt to track the state of the art and recent results in speech recognition, providing a comprehensive overview of the field.

### Overview: This comprehensive guide provides a step-by-step approach to building a Retrieval-Augmented Generation (RAG) based LLM application for production. It covers the development, scaling, and optimization of the major components of the application, including load, chunk, embedding, indexing, serving, and LLM hybrid routing. The guide also discusses the impacts of LLM applications on products and shares insights from Anyscale Endpoints. ### Key Features: - 🚀 Scalable load, chunk, embed, index, and serve components. - 🚀 Evaluate different configurations for optimal performance. - 🚀 Implement LLM hybrid routing for OSS and closed LLMs. - 🚀 Serve the application in a highly scalable and available manner. - 🚀 Share the impacts of LLM applications on products. ### Tech Stack: - Jupyter Notebook ### Why It Matters: This guide is essential for developers looking to build and deploy LLM applications that leverage RAG, ensuring high performance and scalability. It provides practical insights into integrating LLMs with other technologies and optimizing the entire application stack. ### Repository: - Repository: ray-project/llm-applications - GitHub Stars: 1862 - GitHub Description: A comprehensive guide to building RAG-based LLM applications for production. ### Setup: - **API Keys**: Access OpenAI and Anyscale Endpoints. - **Compute**: Use a GPU cluster on Anyscale or set up a local environment. - **Data**: Data is already available at `/efs/shared_storage/goku/docs.ray.io

### Overview: Babysitter is a powerful tool designed to enforce obedience and manage complex workflows among agentic workforces. It ensures every step is executed deterministically, preventing hallucination and maintaining an immutable record of every decision. Babysitter supports multiple AI coding harnesses, making it highly versatile and compatible with various environments. ### Key Features: - **Deterministic Execution**: Ensures every step is executed exactly as planned. - **Immutable Journal**: Records every decision in an unchangeable log. - **Quality Gates**: Requires human approval at breakpoints to ensure quality. - **Adapters Support**: Runs across 12 supported AI coding harnesses, including Claude Code. ### Tech Stack: Node.js 22.13.0+ with the `adapters` CLI. ### Why It Matters: Babysitter simplifies the management of complex workflows, ensuring that workforces follow defined processes without the risk of hallucination or human error. It provides a robust, deterministic, and auditable way to manage tasks and workflows, making it ideal for environments where precision and reliability are crucial. ### Repository: - **Repository**: a5c-ai/babysitter - **Stars**: 1862 - **Description**: Babysitter enforces obedience on agentic workforces and enables them to manage extremely complex tasks and workflows through deterministic, hallucination-free self-orchestration. ### Installation: - **Node.js**: Version 20.0.0+ (22.x LTS recommended). - **Harness**: Any of the 1

### Overview: DPM-Solver is an official code for the paper "DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps" (Neurips 2022 Oral). This solver is designed to efficiently generate high-quality samples in just 10 to 20 function evaluations, making it suitable for various diffusion models without the need for further training. ### Key features: - **Fast Convergence Order**: DPM-Solver guarantees high-order convergence, ensuring accurate and efficient sampling. - **Discrete-Time and Continuous-Time Compatibility**: The solver is compatible with both discrete-time and continuous-time diffusion models, making it versatile for various applications. - **Efficient Sampling**: DPM-Solver can generate high-quality samples in just 10 to 20 function evaluations, significantly reducing computational time. ### Tech stack: Python ### Why it matters: DPM-Solver is a powerful tool for generating samples in diffusion models, offering fast and accurate solutions without the need for additional training. Its efficiency makes it an invaluable asset in the field of diffusion probabilistic modeling, particularly for applications requiring rapid sampling. Repository: [LuChengTHU/dpm-solver](https://github.com/LuChengTHU/dpm-solver) GitHub stars: 1859

### Overview: This project is a comprehensive global situational awareness platform that provides real-time mapping of security events, geopolitical developments, and threat indicators. It offers interactive features like real-time event mapping, an event feed, and detailed intelligence dossiers on actors, along with a military bases layer and a tabbed interface for historical and current conflicts. ### Key Features: - **Real-Time Event Mapping**: Plots breaking news events on a world map with color-coded threat levels. - **Interactive Mapbox Map**: Dark-themed map with clustering, heatmap visualization, and smooth navigation. - **Event Feed**: Real-time filterable feed of global events with category and threat level filters. - **Intel Dossiers**: Build intelligence dossiers on any actor with deep research reports, CSV data exports, and PowerPoint briefings. - **Alert System**: Configure keyword and region-based alerts with real-time notifications. ### Tech Stack: - **Framework**: Next.js 16 (App Router) - **Map**: Mapbox GL JS + react-map-gl - **UI**: Tailwind CSS v4 + custom components - **Intelligence API**: valyu-js (search, answer, deep research) - **Schema Validation**: zod - **State Management**: Zustand - **Markdown**: react-markdown + remark-gfm ### Why it Matters: This platform serves as a powerful tool for security analysts, researchers, and policymakers to gain real-time insights into geopolitical events, military activities, and historical conflicts. It helps in understanding the global security landscape, tracking geopolitical tensions, and making

Overview: AI-Blocks is a user-friendly WYSIWYG interface designed to simplify the creation of Machine Learning models, making it accessible to anyone. Key features: - Drag-and-drop interface for easy model creation - Integration with Python scripts for variable manipulation - Direct execution on the editor or export to standalone TensorFlow script - Run-time execution of models Tech stack: JavaScript Why it matters: AI-Blocks offers a simple, intuitive way to create and run Machine Learning models without requiring deep knowledge of programming languages, making it accessible to a wide audience.

### Overview: Hope Agent is a cross-device desktop AI assistant that learns from your interactions, dynamically orchestrates tasks, and runs autonomously. It seamlessly transitions between devices and continuously improves based on usage, making it a powerful tool for productivity and personalization. ### Key Features: - **Dynamic Workflow**: Orchestrates tasks across multiple devices dynamically. - **Goal-Oriented**: Pursues goals autonomously, adapting to changing contexts. - **Memory**: Remembers your preferences and interactions, improving over time. - **Headless Deployment**: Runs without a frontend, suitable for services or cloud environments. ### Tech Stack: Rust ### Why It Matters: Hope Agent is a pioneering tool that combines the power of AI with the simplicity of a desktop assistant, making it a valuable asset for anyone looking to enhance their productivity and personal efficiency. Its ability to learn and adapt over time makes it a versatile and reliable companion for both personal and professional use. Repository: shiwenwen/hope-agent GitHub Stars: 1858

### Overview: This repository is a curated list of awesome embedding models tutorials, projects, and communities. It serves as a comprehensive resource for developers looking to understand and utilize various embedding techniques in their projects. ### Key Features: - **Word Embeddings**: Includes popular models like Word2Vec, GloVe, FastText, and their derivations. - **Language Models**: Covers advanced models such as BERT, Deep contextualized word representations, and others. - **Embedding Enhancement**: Discusses methods for enhancing word embeddings, including semantic lexicons and dependency-based embeddings. - **Comparing Methods**: Provides insights into comparing count-based and predict-based methods for word embeddings. ### Tech Stack: - Jupyter Notebook ### Why It Matters: This repository is essential for developers interested in natural language processing, machine learning, and information retrieval. It offers a wealth of resources to help developers understand and implement advanced embedding techniques effectively.

### Overview: This repository is a comprehensive resource list for low light image enhancement, featuring datasets, methods, papers, and more. It aims to facilitate the development of new techniques and solutions for low light imaging tasks. ### Key Features: - **Datasets**: Includes datasets such as SID, ExDARK, LOL, SICE, MIT-Adobe FiveK, DID, and DPED, covering various low-light scenarios and object classes. - **Methods**: Covers learning-based methods, HE-based methods, Retinex-based methods, and other techniques. - **Metrics**: Provides metrics for evaluating low light image enhancement techniques. ### Tech Stack: MATLAB ### Why It Matters: This list aims to provide a valuable resource for researchers and developers working on low light image enhancement, helping to advance the field and improve the quality of low light imaging applications.

### Overview: Gen.jl is a general-purpose probabilistic programming system that allows for programmable inference, embedded in the Julia programming language. It supports a variety of inference methods, including Sequential Monte Carlo, variational inference, and Markov Chain Monte Carlo, among others. The system is designed to be intuitive and expressive, making it suitable for a wide range of applications in probabilistic programming. ### Key Features: - **Multi-paradigm Inference**: Supports multiple inference methods including Sequential Monte Carlo, variational inference, and MCMC. - **Gradient-Based Training**: Enables training of generative models using gradient-based optimization techniques. - **Modeling Language**: Provides an expressive and intuitive modeling language for writing and composing probabilistic programs. - **Programmable Algorithms**: Allows for the customization of inference algorithms without needing to understand the underlying mathematical concepts. - **Structure Learning**: Supports Bayesian structure learning using involutive MCMC and Specialized Modeling Constructs. - **Specialized Constructs**: Speeds up inference by supporting incremental computation with specialized constructs. - **Custom Implementations**: Provides APIs for implementing custom generative models, distributions, and gradients. ### Tech Stack: Julia ### Why It Matters: Gen.jl is designed to be a versatile tool for probabilistic programming, making it accessible and powerful for a wide range of applications in machine learning, statistics, and other fields that involve probabilistic modeling. Its ability to support multiple inference methods and its intuitive modeling language make it a valuable resource for researchers and practitioners alike. ### Repository: - **Repository**: [probcomp/Gen.j

### Overview: Spago is a self-contained Machine Learning library written in Go, designed to support various neural architectures in Natural Language Processing (NLP). ### Key features: - Automatic differentiation via dynamic define-by-run execution - Feed-forward layers (Linear, Highway, Convolution...) - Recurrent layers (LSTM, GRU, BiLSTM...) - Attention layers (Self-Attention, Multi-Head Attention...) - Gradient descent optimizers (Adam, RAdam, RMS-Prop, AdaGrad, SGD) - Gob compatible neural models for serialization ### Tech stack: Go ### Why it matters: Spago is an extraordinary project that has significantly contributed to the development of minimalist ML frameworks in Go, focusing on clean coding and eliminating the dependency on Python in production. It has been instrumental in powering several of the author's projects in challenging production environments. The author is deeply grateful for the journey Spago has taken them on and the community that has supported it.

Overview: This repository is a curated list of awesome semi-supervised learning resources, inspired by other popular repositories like [awesome-deep-vision](https://github.com/kjw0612/awesome-deep-vision) and [awesome-deep-learning-papers](https://github.com/terryum/awesome-deep-learning-papers). It aims to provide a comprehensive overview of semi-supervised learning methods, techniques, and resources for developers. Key features: - Papers: A collection of semi-supervised learning papers, methods, and resources. - Repositories: Links to repositories where you can find implementations and code for various semi-supervised learning techniques. - Resources: Additional resources such as blog posts, tutorials, and more. - Community: A way to contribute to the list by adding new papers or improving existing ones. Tech stack: The repository is written in Markdown and uses GitHub's repository features. Why it matters: Semi-supervised learning is a crucial area in machine learning, especially in scenarios where labeled data is scarce or expensive to obtain. This repository serves as a valuable resource for developers looking to improve the accuracy of their models while reducing the need for human annotation. Repository: yassouali/awesome-semi-supervised-learning GitHub stars: 1853 GitHub description: 😎 An up-to-date & curated list of awesome semi-supervised learning papers, methods & resources.