Projects

Overview: This repository contains examples for customers to get started using the Amazon Bedrock Service, including tips for prompt engineering, ways to implement Generative AI Agents, and examples for various use cases such as Retrieval-Augmented Generation (RAG). Key features: - Prompt Engineering - Agents - Custom Model Import - Multimodal - Generative AI Use cases - Retrieval Augmented Generation (RAG) - Responsible AI - Workshop - POC to Prod - Embeddings - Observability & Evaluation Tech stack: Jupyter Notebook Why it matters: This repository provides pre-built examples to help customers quickly get started with Amazon Bedrock, including tips for effective prompt engineering, ways to implement Generative AI Agents, and various use cases such as Retrieval-Augmented Generation (RAG) and Responsible AI. Repository: aws-samples/amazon-bedrock-samples GitHub stars: 1524 GitHub description: This repository contains examples for all available foundational models in the Amazon Bedrock service.

Overview: This project streamlines the development process by leveraging AI to enhance the speed and efficiency of UI creation, allowing developers to focus on more critical tasks. Key features: - AI-driven UI design - Real-time collaboration - Automated testing - Improved code quality Tech stack: Python Why it matters: This tool accelerates the development cycle, enabling teams to deliver projects faster and with higher quality, thereby reducing development time and costs.

### Overview: WebBrain is an open-source AI browser extension designed to enhance your browsing experience by integrating an AI agent directly into your browser tabs. It allows you to interact with web pages, automate tasks, and run multi-step workflows using various language models, including local LLMs and managed servers. ### Key Features: - **Ask Mode**: Reads the page, answers questions, and fetches URLs. - **Act Mode**: Clicks, types, navigates, uploads, downloads, fills forms. - **Dev Mode**: Adds page source, styles, console, network, and reversible page edits. ### Tech Stack: JavaScript ### Why it Matters: WebBrain is a powerful tool that enhances your browsing experience by integrating an AI agent directly into your browser tabs. It allows you to interact with web pages, automate tasks, and run multi-step workflows using various language models, including local LLMs and managed servers. This makes it a valuable addition to any developer's toolkit, enhancing productivity and efficiency. ### Repository: - **Repository**: webbrain-one/webbrain - **GitHub Stars**: 1522 - **GitHub Description**: Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠 ### Installation: - **Chrome**: Install from the [Chrome Web Store](https://chromewebstore.google.com/detail/webbrain/ljhijonmfahplgbbacgcfnaihbjljhhb). - **Firefox**: Install from [Firefox Add-ons](https://addons.mozilla.org/firefox/addon/webbrain/). - **

### Overview: OpenContracts is an open-source document intelligence platform designed for builders and hackers, providing a self-hosted, MIT-licensed solution for teams working at scale. ### Key features: - **GraphQL + REST API**: Programmable citation graph for your apps. - **Model Context Protocol**: Server for your agents. - **React UI**: Team collaboration and management. ### Tech stack: Python ### Why it matters: OpenContracts offers a comprehensive, open-source document intelligence platform that can be easily integrated into various applications, making it a valuable tool for legal professionals and developers alike.

### Overview: Unblink V2 is an AI-powered camera monitoring application built with a relay-node architecture and real-time video processing capabilities. It utilizes the Qwen3-VL Vision Language Model for frame analysis and summarization, offering a chat interface for natural language interaction with camera feeds and advanced video search features. ### Key Features: - **Vision Language Model**: Qwen3-VL for frame analysis and summarization. - **Chat Interface**: Natural language interaction with your camera feeds. - **Video Search**: Search through recorded frames using natural language. ### Tech Stack: Go ### Why it Matters: Unblink V2 provides a robust solution for camera monitoring, leveraging advanced AI and real-time video processing to enhance security and operational efficiency. Its architecture, combining public and private nodes, ensures seamless integration and scalability, making it ideal for both small and large-scale deployments.

Overview: TheoremExplainAgent (TEA) is an AI system designed to generate visual explanations of theorems using video, aiming to enhance the understanding and uncovering of reasoning flaws in language models like LLMs. Key features: - Generates long-form Manim videos to explain theorems - Incorporates multimodal explanations for better understanding - Identifies and highlights reasoning flaws that text alone often misses Tech stack: Python Why it matters: TheoremExplainAgent provides a novel and effective method for enhancing the clarity and understanding of complex mathematical concepts, making it valuable for researchers and educators alike. Repository: TIGER-AI-Lab/TheoremExplainAgent Primary language: Python GitHub stars: 1514 GitHub description: Official Repo for "TheoremExplainAgent: Towards Video-based Multimodal Explanations for LLM Theorem Understanding" [ACL 2025 oral] README excerpt:

Overview: Kodus is an open-source, self-hosted AI code review tool that allows developers to review pull requests on various platforms, including GitHub, GitLab, Bitbucket, Azure DevOps, and Forgejo. It supports multiple LLMs and offers a flexible model agnostic approach, enabling teams to choose their preferred language model. Key features: - Supports multiple LLMs (OpenAI, Anthropic, Google Gemini, etc.) - Zero markup on LLM costs - Learns from your context - You set the rules - Privacy & security: source code not used for training, data encrypted, self-hosted runners supported - Native Git workflow - CLI + CI/CD ready - Operational impact: track technical debt and delivery metrics Tech stack: TypeScript Why it matters: Kodus provides a flexible and secure way for teams to review code using their preferred LLM, while maintaining privacy and reducing costs. Its native Git workflow and CLI capabilities make it easy to integrate with existing CI/CD pipelines, ensuring high-quality code reviews without compromising on operational impact.

Overview: gawkbot is an open-source Grok Bot that automates menial work tasks using AI models and builds microapps to manage outcomes, providing a false sense of control. Key features: - Automates manual workflows into microapps - Integrates with 1200+ integrations - Built on sustainable use license - Runs locally on your machine - AI models from various sources - Built-in approval gate for external actions Tech stack: Go Why it matters: gawkbot simplifies complex tasks by automating them into microapps, making them more manageable and reducing the need for manual intervention.

### Overview: This repository is a collection of datasets, papers, and books related to the intersection of deep learning and finance. It serves as an excellent resource for those interested in the latest developments in time series forecasting, natural language processing, graph neural networks, recommendation systems, and finance. ### Key features: - **Stock Prediction**: Comprehensive dataset for stock movement prediction from tweets and historical stock prices. - **Portfolio Selection**: Financial sentence boundary detection dataset for identifying financial sentences in text. - **Risk Management**: Financial phrase boundary detection dataset for financial text segmentation. - **Finance NLP**: Financial QA dataset for financial question answering. - **Blockchain**: Financial sentence boundary detection dataset for blockchain applications. - **Market Maker**: Financial sentence boundary detection dataset for market maker scenarios. - **Others**: Financial sentiment analysis dataset for sentiment analysis in financial news. ### Tech stack: Python, TensorFlow, PyTorch, and other deep learning frameworks. ### Why it matters: The ideas and models presented in these papers can provide valuable insights and practical applications for developers and researchers in the field of deep learning and finance.

### Overview: Multimodal-GPT is a project that combines the power of both visual and language instruction to train a multi-modal chatbot. This project leverages the OpenFlamingo model to create comprehensive datasets and fine-tunes it effectively, making it a valuable tool for developers looking to enhance their chatbot capabilities. ### Key Features: - **Supports Various Vision and Language Instruction Data**: The project includes datasets for Visual Question Answering (VQA), Image Captioning, Visual Reasoning, Text OCR, and Visual Dialogue, ensuring a wide range of applications. - **Parameter Efficient Fine-Tuning with LoRA**: The model uses LoRA (Low-Rank Adaptation) for efficient parameter tuning, making it suitable for fine-tuning on limited computational resources. - **Tuning Vision and Language at the Same Time**: The project effectively combines the strengths of both vision and language models, enhancing the overall performance of the chatbot. ### Tech Stack: Python ### Why It Matters: Multimodal-GPT offers a powerful solution for developers looking to create more versatile and intelligent chatbots. By combining visual and language instruction, it addresses the limitations of traditional chatbots and opens up new possibilities in natural language processing and computer vision. Repository: open-mmlab/Multimodal-GPT GitHub Stars: 1511

### Overview: 这本书是《深度学习与计算机视觉：实例入门》的配套代码，提供了一系列入门级别的深度学习和计算机视觉项目示例，涵盖图像分类、物体检测、卷积神经网络、迁移学习等主题。 ### Key Features: - 实例化的深度学习和计算机视觉项目 - 包含图像分类、物体检测、卷积神经网络、迁移学习等主题 - 使用Python语言编写代码 - 提供彩色图表和电子版彩色图表 - 包含勘误表和错误报告 ### Tech Stack: Python ### Why it Matters: 这本书是深度学习和计算机视觉领域的入门级指南，适合初学者快速上手。通过实际项目和示例代码，读者可以轻松理解并应用深度学习和计算机视觉技术。

Overview: DDColor is a state-of-the-art image colorization model that achieves remarkable performance on historical black and white images, providing vivid and natural colorization. It has been successfully applied to recoloring anime landscapes, transforming them into realistic real-life styles. Key features: - Automatically colorizes historical black and white images - Recolors landscapes from anime games - Supports inference via Hugging Face, ModelScope, and Replicate Tech stack: Python, PyTorch, Hugging Face, ModelScope, Replicate Why it matters: DDColor is a significant advancement in the field of image colorization, offering high-quality colorization for historical and anime scenes, making it highly valuable for both artistic and technical applications. Repository: piddnad/DDColor Primary language: Python GitHub stars: 1509

### Overview PyCM is a multi-class confusion matrix library written in Python that supports both input data vectors and direct matrix, and is a valuable tool for post-classification model evaluation. It provides a broad array of metrics for predictive models and accurate evaluation of a wide variety of classifiers. ### Key Features - **Supports Direct Matrix and Input Data Vectors**: PyCM can handle both direct confusion matrix data and input data vectors, making it versatile for various applications. - **Comprehensive Metrics**: It supports most classes and provides overall statistics parameters, ensuring comprehensive evaluation. - **Swiss-Army Knife Approach**: PyCM is designed to be a versatile tool, suitable for data scientists and researchers needing a wide range of metrics for their models. ### Tech Stack - Python ### Why It Matters PyCM is essential for data scientists and researchers who need a robust and versatile tool for evaluating machine learning models. Its ability to handle both direct matrix and input data vectors makes it highly adaptable to different datasets and evaluation needs.

### Overview: AutoGroq is a revolutionary tool that automates the creation of expert AI teams tailored to your project needs, streamlining the interaction with AI assistants like Autogen. ### Key Features: - **Dynamic Expert Agent Generation**: Automatically creates specialized AI agents for various domains or topics. - **Dynamic Workflow Generation**: Instantly sets up a custom team of experts for your project. - **Natural Conversation Flow**: Engages in context-aware conversations with AI agents. - **Code Snippet Extraction**: Presents code snippets for easy reference. - **Flexible Agent Management**: Customize and modify your team as needed. - **Advanced Prompt Rephrasing**: Enhances clarity and accuracy with natural language processing. - **Bulk File Upload to Autogen**: Imports multiple agents, skills, and workflows. - **Support for Multiple LLMs**: Compatible with Groq, ChatGPT, Ollama, and more. - **Skill Integration**: Adds custom skills for enhanced capabilities. ### Tech Stack: Python ### Why It Matters: AutoGroq simplifies the process of creating and managing AI teams, making it easier and more efficient to tackle any project or question with the right support.

### Overview: 这是一个基本包，封装了大多数NLP项目中常用工具，适用于Java环境。 ### Key features: - 词语标准化 - 简繁体转换 - 汉字转拼音 - 理解和处理HTML标签 - Viterbi算法增强 - 指纹去重 - 基于内存的搜索提示 - WordWeight词频统计,词idf统计,词类别相关度统计 ### Tech stack: Java ### Why it matters: 这个项目为Java开发者提供了基本的NLP工具，简化了开发过程，提高了代码质量和效率。

### Overview: LLM Vision is a Home Assistant integration that leverages multimodal large language models to analyze images, videos, live camera feeds, and Frigate events. It provides a timeline of camera events, allowing users to view and interact with them on their dashboard. ### Key Features: - **Multimodal Analysis**: Analyzes images, videos, live camera feeds, and Frigate events. - **Dynamic Timeline**: Keeps a timeline of camera events, making it easy to track and understand events. - **AI-Enhanced Insights**: Answers questions and provides descriptions based on prompts. - **Event Memory**: Remembers people, pets, and objects. - **Sensors Integration**: Seamlessly updates sensors based on data extracted from camera streams, images, or videos. ### Tech Stack: Python ### Why It Matters: LLM Vision offers a powerful tool for visual intelligence in the home, enhancing security, monitoring, and interaction with smart home devices. It simplifies the process of integrating AI into Home Assistant, making it easier for users to manage and understand their smart home environment. ### Repository: Repository: valentinfrlch/ha-llmvision GitHub Stars: 1503 GitHub Description: Visual intelligence for your home.

Overview: MMEngine is a comprehensive library for training deep learning models using PyTorch. It serves as the core engine for OpenMMLab's research projects and is designed to be flexible and reusable across different applications. Key features: - Support for custom `artifact_location` in MLflowVisBackend - Enable `exclude_frozen_parameters` for `DeepSpeedEngine._zero3_consolidated_16bit_state_dict` Tech stack: Python Why it matters: MMEngine is essential for researchers and developers working on deep learning models, providing a robust and versatile framework that can be easily integrated into various projects.

### Overview: Finetuner simplifies the process of fine-tuning pre-trained models for neural search tasks, making it accessible and efficient. ### Key Features: - Streamlines the workflow for task-oriented fine-tuning. - Handles all complexity and infrastructure in the cloud. - Enhances the performance of pre-trained models, improving semantic search, visual similarity, and cross-modal text->image search. - Supports 40+ mainstream loss functions, 10+ optimizers, layer pruning, weight freezing, dimensionality reduction, hard-negative mining, cross-modal models, and distributed training. - Easy to use, with all-in-cloud training and management on Jina AI Cloud. - Ideal for low budget, high expectations scenarios, delivering considerable improvements with minimal training samples and short runtime. ### Tech Stack: Python ### Why it Matters: Finetuner makes fine-tuning easier and faster, streamlining the process and handling all the complexity and infrastructure in the cloud, making it accessible and efficient for neural search tasks.

Overview: OctoTools is an agentic framework designed to enhance complex reasoning capabilities through extensible tools. Key features: - Support for multiple LLM engines, including Ollama, LiteLLM, and vLLM. - Backend support for Azure OpenAI. - Backend support for a broader range of LLM engines. - A comprehensive tutorial available on YouTube. Tech stack: Python Why it matters: OctoTools provides a powerful framework for enhancing complex reasoning capabilities, making it easier to integrate with various LLM engines and backend services.

### Overview: Arkon is a self-hosted, enterprise-grade knowledge management layer that integrates Claude and other large language models (LLMs) using the Model Context Protocol (MCP). It provides a centralized MCP Server for secure, permission-scoped knowledge integration, ensuring consistent and contextually relevant AI interactions across the organization. ### Key Features: - **Intelligent Knowledge Wiki - the MRP Pipeline:** Compiles documents into a coherent wiki of interlinked pages, ensuring consistent context and traceable claims. - **Wiki Browser & Knowledge Graph:** Three-panel layout for page tree, content viewer, and knowledge graph visualization, with full-text and semantic search capabilities. - **Department & Global Scopes:** Precise context boundaries with isolated department-level wikis and a global scope for company-wide SOPs and policies. - **Fine-Grained RBAC:** Role-based access control with built-in roles and granular permissions, including audit logs for privileged actions. - **MCP Server for Claude & Other AI Clients:** Connects Claude Desktop or Claude.ai via OAuth 2.1 + PKCE, exposing wiki tools and secure knowledge integration. ### Tech Stack: Python ### Why it Matters: Arkon provides a centralized, secure, and contextually relevant knowledge management solution for teams, ensuring consistent and contextually relevant AI interactions across the organization.