ai-by-hand-excel is a resource repository that facilitates hands-on practice of AI technology using Excel. It provides a series of Excel files that allow users to manually execute and understand key operations of AI models, such as Softmax, LeakyReLU, Backpropagation, and Transformers. This repository is suitable for beginners and educators looking to gain a deeper understanding of the internal mechanisms of AI models, helping them enhance their grasp of AI technology through practical operation.

Local Knowledge Graph is a Flask-based web application that uses a local Llama language model to process user queries, generate step-by-step reasoning, and visualize the thought process in an interactive knowledge graph format. It can also find and display relevant questions and answers based on semantic similarity. Key advantages of this application include real-time visualization of the reasoning process, dynamic knowledge graph representation, computation and display of the strongest reasoning paths, and related Q&A based on semantic similarity.

AI Timeline is an open-source project that records significant technological milestones in the field of artificial intelligence. It meticulously chronicles key developments in AI technologies, including text-to-image, text-to-video, and large language models. The project is built with Vue and TypeScript, providing a platform for AI enthusiasts and researchers to explore the history and trends of AI.

IncarnaMind is an open-source project aimed at enabling conversational interactions with personal documents (PDF, TXT) using large language models (LLMs) such as GPT, Claude, and local open-source LLMs. The project enhances query efficiency and improves the accuracy of LLMs through a sliding window chunking mechanism and integrated retrievers. It supports multi-document conversational Q&A, overcoming the limitations of single-document interaction while being compatible with various file formats and LLM models.

GraphRAG-Ollama-UI is a local adaptation version of Microsoft's GraphRAG, supporting local model support using Ollama. A interactive user interface is provided through the Gradio UI, making it more convenient for users to manage data, run queries and visualize results. The main advantages of this model include local model support, cost-effectiveness, interactive user interface, real-time graph visualization, file management, settings management, output exploration, and logging.

MaxKB is an enterprise-grade knowledge base question-answering system powered by LLMs. It aims to be the ultimate brain for businesses. Features include document upload, automatic online document crawling, and a smart question-answering experience. It also supports quick embedding into third-party business systems. The tech stack includes Vue.js, Python/Django, Langchain, and PostgreSQL/pgvector.

MathPile is a mathematics-centric corpus containing approximately 9.5 billion tokens. It draws mathematical content from textbooks (including lecture notes), arXiv, Wikipedia, ProofWiki, StackExchange, and web pages. It is suitable for K-12, university, graduate-level, and math competition applications. MathPile boasts high data quality and comprehensive data documentation to enhance transparency and provide users with flexible data utilization capabilities. MathPile adheres to the BY-NC-SA 4.0 license and plans to release a commercially available version soon.

DocGraphLM is a document graph language model for information extraction and question answering. It employs advanced vision-rich document understanding techniques, combining pre-trained language models and graph semantics. Its uniqueness lies in proposing a joint encoder architecture to represent documents and a novel link prediction method to reconstruct the document graph. DocGraphLM predicts the direction and distance between nodes through a convergent joint loss function, prioritizing neighborhood restoration and minimizing the weight of remote node detection. Experiments on three SotA datasets demonstrate that incorporating graph features consistently improves performance in information extraction and question-answering tasks. Furthermore, we report that employing graph features accelerates convergence during training, despite these features being constructed solely through link prediction.

ai-by-hand-excel is a resource repository that facilitates hands-on practice of AI technology using Excel. It provides a series of Excel files that allow users to manually execute and understand key operations of AI models, such as Softmax, LeakyReLU, Backpropagation, and Transformers. This repository is suitable for beginners and educators looking to gain a deeper understanding of the internal mechanisms of AI models, helping them enhance their grasp of AI technology through practical operation.

Local Knowledge Graph is a Flask-based web application that uses a local Llama language model to process user queries, generate step-by-step reasoning, and visualize the thought process in an interactive knowledge graph format. It can also find and display relevant questions and answers based on semantic similarity. Key advantages of this application include real-time visualization of the reasoning process, dynamic knowledge graph representation, computation and display of the strongest reasoning paths, and related Q&A based on semantic similarity.

AI Timeline is an open-source project that records significant technological milestones in the field of artificial intelligence. It meticulously chronicles key developments in AI technologies, including text-to-image, text-to-video, and large language models. The project is built with Vue and TypeScript, providing a platform for AI enthusiasts and researchers to explore the history and trends of AI.

IncarnaMind is an open-source project aimed at enabling conversational interactions with personal documents (PDF, TXT) using large language models (LLMs) such as GPT, Claude, and local open-source LLMs. The project enhances query efficiency and improves the accuracy of LLMs through a sliding window chunking mechanism and integrated retrievers. It supports multi-document conversational Q&A, overcoming the limitations of single-document interaction while being compatible with various file formats and LLM models.

GraphRAG-Ollama-UI is a local adaptation version of Microsoft's GraphRAG, supporting local model support using Ollama. A interactive user interface is provided through the Gradio UI, making it more convenient for users to manage data, run queries and visualize results. The main advantages of this model include local model support, cost-effectiveness, interactive user interface, real-time graph visualization, file management, settings management, output exploration, and logging.

MaxKB is an enterprise-grade knowledge base question-answering system powered by LLMs. It aims to be the ultimate brain for businesses. Features include document upload, automatic online document crawling, and a smart question-answering experience. It also supports quick embedding into third-party business systems. The tech stack includes Vue.js, Python/Django, Langchain, and PostgreSQL/pgvector.

MathPile is a mathematics-centric corpus containing approximately 9.5 billion tokens. It draws mathematical content from textbooks (including lecture notes), arXiv, Wikipedia, ProofWiki, StackExchange, and web pages. It is suitable for K-12, university, graduate-level, and math competition applications. MathPile boasts high data quality and comprehensive data documentation to enhance transparency and provide users with flexible data utilization capabilities. MathPile adheres to the BY-NC-SA 4.0 license and plans to release a commercially available version soon.

DocGraphLM is a document graph language model for information extraction and question answering. It employs advanced vision-rich document understanding techniques, combining pre-trained language models and graph semantics. Its uniqueness lies in proposing a joint encoder architecture to represent documents and a novel link prediction method to reconstruct the document graph. DocGraphLM predicts the direction and distance between nodes through a convergent joint loss function, prioritizing neighborhood restoration and minimizing the weight of remote node detection. Experiments on three SotA datasets demonstrate that incorporating graph features consistently improves performance in information extraction and question-answering tasks. Furthermore, we report that employing graph features accelerates convergence during training, despite these features being constructed solely through link prediction.