Projects

### Overview: Vestige is a local, deterministic safety kernel designed for AI agents. It ensures that every memory write and agent action is recorded in Strata, an append-only log, allowing for precise tracking and review of actions. ### Key Features: - **Guard against force pushes, dropped tables, and production deploys** using Operator Lite. - **Backfill mechanism** that remembers what matters, allowing agents to see and correct their past actions. - **Strata log** for precise memory tracking and retrieval. - **Operator Lite** guard that stops unapproved actions and provides a weekly report. ### Tech Stack: Rust ### Why It Matters: Vestige ensures that AI agents are responsible and accountable for their actions, preventing accidental or malicious changes to their environment. It provides a clear and verifiable history of agent behavior, making it easier to understand and debug complex systems. ### Repository: GitHub Repository: samvallad33/vestige ### GitHub Stars: 656 ### Primary Language: Rust ### Binary Platforms: - macOS - Linux ### Release Targets: - 6 ### License: AGPL-3.0 ### Quick Start: 1. Install the binaries or download them from Releases. 2. Run `npx @vestige/init` to choose between Memory or Memory + Guard. 3. Restart your AI client and say: *"Tell me something about a friend."* Ask about them later. ### Backfill Mechanism: - **Retroactive Salience Backfill**: When a new memory makes an old one

### Overview: SpecAugment is a speech data augmentation method that directly processes spectrograms using Tensorflow and PyTorch. It was introduced by Google Brain and is currently under the Apache 2.0 license. This implementation is designed to enhance the robustness of speech recognition models by applying specific transformations to the spectrograms. ### Key Features: - **Tensorflow & Pytorch Implementation**: Utilizes TensorFlow and PyTorch for efficient processing. - **SpecAugment Method**: Directly modifies spectrograms by warping them in the time direction, masking consecutive frequency channels, and masking blocks of utterances in time. - **Audio Libraries**: Requires Python 3 and audio libraries such as Librosa to be installed. ### Tech Stack: - Python 3 - TensorFlow - PyTorch ### Why It Matters: SpecAugment is a powerful technique that can significantly improve the performance of speech recognition models by mitigating the effects of time and frequency masking. It is widely used in the field of speech processing and has been shown to enhance the accuracy of speech recognition systems. ### Repository: - **Repository**: DemisEom/SpecAugment - **Stars**: 655 - **Description**: A implementation of SpecAugment that speech data augmentation method which directly process the spectrogram with Tensorflow & Pytorch, introduced by Google Brain[1]. This is currently under the Apache 2.0, Please feel free to use for your project. Enjoy! ### How to Use: 1. **Install Requirements**: ```bash pip3 install

Overview: This repository contains the Deep Learning Specialization course materials from Andrew Ng's Coursera course, including IPYNB notebooks, slides, and pre-trained models. Key features: - IPYNB notebooks and slides associated with the course - All datafiles added - You can download them and practice offline - All slides added - Pre-trained models like ResNet available for download Tech stack: Python, Jupyter Notebook, and other relevant libraries. Why it matters: This repository provides a comprehensive resource for anyone interested in deep learning, offering a practical way to learn and practice the material without violating Coursera's terms of use.

Overview: 聚宝盆(Cornucopia)是一个基于LLaMA系列模型进行指令微调的开源项目，旨在提供中文金融领域的微调模型和高效轻量化的训练框架。 Key features: - 指令微调的LLaMA模型 - 基于中文金融知识的指令数据集 - 提供多种Prompt形式和多任务形式的指令数据集 - 支持多种训练序列长度和版本 Tech stack: Python Why it matters: 该项目为金融领域的开发者提供了高质量的微调模型和训练框架，有助于提升模型在金融领域的应用效果，同时提供了丰富的指令数据集，增强了模型的泛化能力和多任务处理能力。

Overview: SmythOS Studio is an open-source visual AI agent builder and deployable runtime stack designed for developers. It allows users to create and manage intelligent AI agents without writing code, offering an intuitive drag-and-drop workspace and the flexibility to extend with custom code. Key features: - Visual Agent Building: Create complex agent workflows using a drag-and-drop interface. - No-Code to Pro-Code: Combine visual building with the ability to extend with custom code. - Open Architecture: Build once, deploy anywhere with full control over infrastructure. Tech stack: TypeScript, React, Node.js, and various libraries for AI integration. Why it matters: SmythOS Studio simplifies AI agent development, making it accessible to both business analysts and developers, while offering the flexibility to extend with custom code and integrate with various AI components. The visual interface and real-time testing capabilities make it an ideal choice for rapid development and deployment.

Overview: This project is a PyTorch implementation of the ICRA 2019 paper "Self-supervised Sparse-to-Dense: Self-supervised Depth Completion from LiDAR and Monocular Camera" by Fangchang Ma, Guilherme Venturelli Cavalheiro, and Sertac Karaman. The code is trained on the KITTI dataset and is designed to complete depth maps from LiDAR and monocular camera inputs. Key features: - Self-supervised training framework - Incorporates photometric loss and sparse depth loss - Uses smoothness loss to improve depth completion accuracy - Supports both supervised and self-supervised training modes Tech stack: Python Why it matters: This project is significant for its contribution to the field of computer vision and depth completion, offering a robust method for depth map generation from sparse and monocular data. The self-supervised training approach allows for the completion of depth maps without the need for ground truth, making it particularly useful for applications where ground truth data is not available. Repository: fangchangma/self-supervised-depth-completion GitHub stars: 654

Overview: kooky is a minimal modern terminal designed for AI coding, featuring sidebar workspaces, horizontal/vertical split panes, one-click agent launches, and live workspace state with easy switching. Key features: - Sidebar workspaces with three-state collapse - Dragging sidebar edges to adjust width - Split panes and windows - One-click AI agent sessions - Workspace color tagging - Agent command history and tool pills Tech stack: Swift Why it matters: kooky simplifies AI coding by providing a clean, minimal interface and efficient workspace management, making it ideal for developers working with AI tools.

### Overview: MultiBench is a comprehensive benchmark for multimodal representation learning, covering 15 datasets, 10 modalities, 20 prediction tasks, and 6 research areas. It aims to accelerate progress in understudied modalities and tasks, ensuring real-world robustness. ### Key Features: - **15 Datasets**: A diverse set of multimodal datasets. - **10 Modalities**: Multiple types of data sources integrated. - **20 Prediction Tasks**: A wide range of tasks for evaluation. - **6 Research Areas**: Focus on generalization, complexity, robustness, and more. - **Automated Pipeline**: Simplifies the machine learning pipeline for easy experimentation. ### Tech Stack: Python, TensorFlow, PyTorch, and other relevant tools. ### Why it Matters: MultiBench provides a standardized benchmark for multimodal learning, facilitating research progress and real-world application development.

Overview: This project is an implementation of YOLOv7 object tracking using PyTorch, OpenCV, and Sort tracking. Key features: - YOLOv7 Object Tracking - Support for YOLOv9, YOLOv10, YOLOv11, YOLOv12, and YOLOv13 - Pretrained YOLOv7 weights available for download Tech stack: Python Why it matters: This project provides a practical application of YOLOv7 for object tracking, showcasing its capabilities in real-time tracking scenarios.

### Overview: This repository, **Awesome-LLM-in-Social-Science**, is a curated collection of papers that evaluate, align, and enhance Large Language Models (LLMs) from a perspective of Social Science. It aims to provide a comprehensive overview of the emerging social science of LLMs, focusing on evaluations, datasets, research, and tools. ### Key Features: - **Evaluation of LLMs**: Papers that systematically review and evaluate LLMs in the context of Social Science. - **Alignment of LLMs**: Papers that align LLMs with Social Science principles and values. - **Research and Tools**: Papers that use LLMs to facilitate research, address issues, and enhance tools in Social Science. - **Surveys**: Papers that compile surveys on various aspects of LLMs in Social Science. - **Perspectives and Positioning**: Papers that offer different perspectives and positions on the use of LLMs in Social Science. - **Datasets**: Papers that contribute datasets related to LLMs in Social Science. - **Simulations**: Papers that simulate scenarios involving LLMs in Social Science. - **Alignment of LLMs with Social Science Principles**: Papers that explore the alignment of LLMs with Social Science values and principles. ### Tech Stack: The repository uses Markdown for its structure and documentation. ### Why It Matters: This collection is crucial for researchers, practitioners, and enthusiasts interested in the intersection of LLMs and Social Science. It provides a valuable resource for understanding the current state of research, identifying gaps, and contributing to the development of

Overview: Character-LLMs is a trainable agent designed for role-playing, allowing characters like Beethoven, Queen Cleopatra, and Julius Caesar to be trained and used with detailed character knowledge and personality. Key features: - Detailed character-related knowledge - Representative character personalities - Trainable agent for role-playing - No additional prompt or reference document needed - Data generation process for experience reconstruction Tech stack: Python Why it matters: Character-LLMs offer a unique approach to role-playing by training agents directly from real experiences and characteristics, making them versatile and contextually accurate.

Overview: `vectordb` is a Pythonic vector database that offers a comprehensive suite of CRUD operations and robust scalability options, including sharding and replication. It is designed to be easily deployable in various environments, from local to on-premise and cloud, and provides a powerful and user-friendly vector database experience. Key features: - CRUD support - Scalability options (sharding, replication) - Easy deployment in local, on-premise, and cloud environments Tech stack: Python Why it matters: `vectordb` is a testament to effective Pythonic design, making it a lean yet powerful solution for all needs. It leverages the powerful retrieval prowess of DocArray and the scalability, reliability, and serving capabilities of Jina, providing a robust and user-friendly vector database experience.

### Overview: WhisperIME is an Android Input Method Editor (IME) that leverages voice recognition technology based on the Whisper engine. It offers a seamless user experience, functioning both as a standalone application and an integrated IME that can be activated via the microphone button in HeliBoard. As a standalone app, Whisper can also translate any supported language to English. ### Key features: - **Voice Recognition**: Utilizes Whisper engine for accurate and efficient voice recognition. - **Offline Operation**: Works entirely offline, ensuring privacy and convenience. - **Multiple Models**: Offers two models: a compact English-only model and a comprehensive multi-lingual model. - **Settings**: Allows users to select and activate Whisper as voice input through the Android settings. - **Translation**: Translates any supported language to English. ### Tech stack: Java ### Why it matters: WhisperIME is designed to comply with Google's requirement that all apps must submit personal identity details to Google starting in 2026/2027. By choosing to not submit these details, WhisperIME will no longer work on certified Android devices after that time, ensuring users can continue using the app without compromising their privacy. ### Donate: - Send a coffee to woheller69@t-online.de - Or via PayPal: https://www.paypal.com/donate?hosted_button_id=XVXQ54LBLZ4AA ### Initial Setup: - Downloads Whisper models (~435 MB) from Hugging Face on first launch. - Enables voice recognition through USB debugging or settings.

### Overview: AgentReview is the first large language model (LLM) based peer review simulation framework designed to explore the academic peer review process. It simulates the entire peer review cycle, including document generation, discussions, and final decisions, to provide insights into the biases and variations in paper decisions due to reviewers' biases. ### Key Features: - **Document Generation**: Generates over 53,800 peer review documents. - **Reviews & Rebuttals**: Accomplishes 10,460 reviews and rebuttals. - **Reviewer-AC Discussions**: Conducts 23,535 discussions. - **Meta-Reviews / Final Decisions**: Achieves 9,414 / 9,414 meta-reviews and final decisions. - **Conferences Covered**: Analyzed 523 submissions across ICCLR 2020, 2021, 2022, and 2023. - **Decision Variation**: Reflects a 37.1% variation in paper decisions due to reviewers' biases. ### Tech Stack: - **Python 3.10+** - **Gradio** for interactive UI ### Why It Matters: AgentReview offers valuable insights into the biases and variations in peer review decisions, providing a valuable resource for improving the design of peer review mechanisms. The study supports sociological theories such as the social influence theory, altruism fatigue, and authority bias, offering a comprehensive understanding of the peer review process.

### Overview: ConvoKit is a toolkit designed to extract conversational features and analyze social phenomena in conversations. It includes several large datasets and scripts demonstrating its use. ### Key Features: - Linguistic Coordination: Measures linguistic influence and relative power between individuals or groups. - Politeness Strategies: Features correlating with politeness and impoliteness. - Expected Conversational Context Framework: Framework for characterizing utterances and terms based on expected conversational context. ### Tech Stack: Python ### Why it Matters: ConvoKit provides a unified interface for extracting conversational features, making it easier to analyze social phenomena in conversations. The toolkit includes several large datasets and scripts demonstrating its use, making it a valuable resource for researchers and developers in the field of conversational analysis.

Overview: This course provides an in-depth understanding of generative AI, including large language models (LLMs) and their training and deployment methods. You'll learn about the transformer architecture, empirical scaling laws, and state-of-the-art tools for optimizing model performance. Key features: - Understanding generative AI fundamentals - Detailed explanation of transformer architecture and LLM training - Application of empirical scaling laws for model optimization - Techniques for fine-tuning and evaluating large language models - Reinforcement learning and its use in LLM-powered applications Tech stack: Jupyter Notebook Why it matters: This course is essential for developers looking to leverage generative AI, enabling them to make informed decisions and build practical applications quickly.

Overview: Free AI-Tips is a weekly newsletter offering bite-sized Python AI for Business tutorials, provided by Business Science. Key features: - Weekly Python AI tutorials - Free access - Updated weekly Tech stack: Python Why it matters: Stay updated with practical AI solutions for business operations, offered for free.

Overview: Holistic 3D reconstruction is a critical area in computer vision, focusing on the accurate and comprehensive reconstruction of 3D scenes from multiple sensorial data sources. This repository serves as a comprehensive list of papers, resources, and datasets dedicated to this field. Key features: - Diverse datasets ranging from scene-level to object-level, including structured 3D models and annotated structures. - Comprehensive tutorials and workshops, such as the ECCV 2020 Workshop on Holistic Scene Structures and the ICCV 2019 Tutorial on Holistic 3D Reconstruction. - Detailed documentation on datasets, including examples and notes for each dataset. Tech stack: The repository is built using Markdown, HTML, and various datasets and resources. Why it matters: Holistic 3D reconstruction is essential for applications in autonomous driving, augmented reality, and virtual reality, where accurate and comprehensive 3D models are crucial for functionality and user experience.

Overview: GPT Home is a project aimed at providing a better alternative to commercial smart home assistants, offering a more intuitive and user-friendly experience. Built on the Raspberry Pi, it utilizes LiteLLM and LangGraph to create a chatbot that can control various home devices. Key features: - Customizable UI - Integration with popular services like Spotify, Philips Hue, and more - Support for alarms, reminders, and general knowledge - Lightning-fast response times Tech stack: Python, LiteLLM, LangGraph Why it matters: GPT Home aims to simplify and enhance the smart home experience, offering a more accessible and efficient alternative to traditional smart home assistants. Repository: judahpaul16/gpt-home Primary language: Python GitHub stars: 652 GitHub description: ChatGPT at home! A better alternative to commercial smart home assistants, built on the Raspberry Pi using LiteLLM and LangGraph.

### Overview: This project is a content-based image retrieval (CBIR) system designed to efficiently search and retrieve similar images based on various features such as color, texture, shape, and deep learning methods. The system evaluates robustness using Mean Average Precision (MAP) and Mean Map (MMAP) metrics. ### Key features: - Implement multiple image features including color histograms, texture descriptors (gabor, daisy, edge, HOG), and deep learning models (VGG19, ResNet152). - Feature fusion techniques to enhance retrieval accuracy. - Dimensionality reduction using random projection to mitigate the curse of dimensionality. - Robustness evaluation using Mean Average Precision (MAP) and Mean Map (MMAP) metrics. ### Tech stack: Python ### Why it matters: This CBIR system is designed to efficiently search and retrieve similar images based on various features, making it useful in applications such as image search, content-based recommendation, and more. The robustness evaluation using Mean Average Precision (MAP) and Mean Map (MMAP) metrics ensures the system's effectiveness across different scenarios. ### Repository: Repository: pochih/CBIR GitHub stars: 652 GitHub description: 🏞 A content-based image retrieval (CBIR) system ### Primary language: Python ### Image retrieval results for "query1 - women dress": #### Query image: <img align='center' style="border-color:gray;border-width:2px;border-style:dashed" src='result/retrieval_result/query1-w