Projects

Overview: 啊哈自然语言处理包提供包括分词、依存句法分析、语义角色标注、自动摘要、语义相似度计算、LDA 主题预测、词云等服务。 Key features: - 分词类 - 标准分词 - NLP 分词 - 分词断句 - 命名实体识别 - 句法分析类 - 依存句法分析 - 摘要类 - TextRank 摘取关键词 - TextRank 摘取关键句和自动摘要 - 语义类 - Word2Vec 词语相似度 - Word2Vec 句子相似度 - 语义角色标注 - LDA 主题预测 - 附加功能 - 简繁转换 - WordCloud 绘制词云 Tech stack: Java Why it matters: 提供了全面的中文自然语言处理能力，适用于多种应用场景，包括文本摘要、主题建模、关键词提取等，提升了文本处理的效率和准确性。 Repository: jsksxs360/AHANLP GitHub stars: 310

Overview: BNLP is a comprehensive natural language processing toolkit designed specifically for Bengali language, offering a variety of functionalities such as tokenization, embedding, POS tagging, named entity recognition, text cleaning, and more. Key features: - Tokenization: Supports basic, NLTK, and SentencePiece tokenizers. - Embeddings: Provides Word2Vec, FastText, and GloVe embeddings. - Part of Speech Tagging: Uses CRF-based POS tagging. - Named Entity Recognition: Supports CRF-based NER. - Text Cleaning: Includes cleaning for Bengali NLP purposes. - Corpus: Offers access to Bengali letters, vowels, punctuation, and stopwords. - Command Line Interface (CLI): Simplifies text processing without Python code. - Pipeline API: Chains multiple NLP operations for complex tasks. Tech stack: Python, Jupyter Notebook, and various NLP models. Why it matters: BNLP streamlines the process of natural language processing for the Bengali language, making it easier to analyze and understand text data in Bengali, which is crucial for applications such as language translation, sentiment analysis, and more. Repository: sagorbrur/bnlp Primary language: Jupyter Notebook GitHub stars: 310 GitHub description: BNLP is a natural language processing toolkit for Bengali Language. Overview: BNLP is a comprehensive natural language processing toolkit designed specifically for Bengali language, offering a variety of functionalities such as tokenization, embedding, POS tagging, named entity recognition, text cleaning, and more.

### Overview: Este curso de Data Science é uma ferramenta educacional que oferece uma visão completa e prática sobre o campo de ciência de dados. ### Key features: - **Machine Learning**: Introdução a técnicas de aprendizado de máquina, incluindo regressão, classificação, e redes neurais. - **Data Visualization**: Práticas de visualização de dados para melhor compreensão e interpretação. - **Python**: Fundamentos em Python para manipulação de dados e implementação de algoritmos. - **Jupyter Notebook**: Prática de desenvolvimento de código e análise de dados em um ambiente colaborativo. - **Projeto Prático**: Implementação de um modelo de Machine Learning em um conjunto de dados real. ### Tech stack: - Python - Jupyter Notebook ### Why it matters: Este curso é essencial para aqueles que desejam ingressar ou aprimorar suas habilidades em ciência de dados, oferecendo uma combinação de teoria prática e prática prática.

Overview: ChatClaw 是一款快速搭建的类 OpenClaw 知识库和桌面管家应用，只需 5 分钟即可在 macOS 和 Windows 上安装和使用。它支持连接多种主流通讯应用，提供强大的技能库和本地知识管理功能，内置 OpenClaw 5000+ 技能，支持多Agent 模式和定时任务，实现高效任务处理和自动化管理。 Key features: - 一键部署至本地电脑 - 沙箱安全，资源占用少 - 支持多种通讯应用连接 - 5000+ 开箱即用技能库 - 多Agent 模式，适应不同场景 - 自动化任务管理，定时抓取和监控 - 本地知识库管理，支持文件分类 - 交互式记忆功能，持续学习和进化 - 企业微信、微信、钉钉、飞书、QQ、WhatsApp 等多渠道远程控制 - 一问多答功能，轻松比较不同专家意见 - 智能侧边栏，快速切换 AI 助手 - 一键启动，方便快捷 Tech stack: Go Why it matters: ChatClaw 是一款功能强大且易于使用的开源本地知识库和桌面管家应用，适用于需要高效任务处理和自动化管理的个人和企业用户。它提供了一站式的解决方案，无需编程即可实现快速部署和高效任务执行，同时具备强大的技能库和本地知识管理功能，满足各种场景需求。

Overview: This repository contains the source code and datasets for "Graph Chain-of-Thought: Augmenting Large Language Models by Reasoning on Graphs" accepted to ACL 2024. Key features: - Graph Chain-of-Thought (Graph-CoT) framework for augmenting LLMs with external textual graphs. - Comprehensive benchmark dataset (GRBench) for evaluating the proposed models. - Real-world graphs from five domains (academic, e-commerce, literature, healthcare, legal) for diverse applications. Tech stack: Python 3.8, PyTorch, Faiss-GPU. Why it matters: This research addresses the limitations of large language models by integrating structured knowledge from graphs, improving their ability to reason and generate coherent, contextually grounded responses.

### Overview: The GFM-RAG is a Graph Foundation Model for Retrieval Augmented Generation that leverages graph neural networks to enhance the reasoning capabilities of a retrieval augmented language model. This model is designed to efficiently and effectively answer questions by combining the power of graph neural networks with the retrieval capabilities of a large language model. ### Key Features: - **Graph Foundation Model (GFM)**: A graph neural network-based retriever that can reason over the graph-index. - **Universal Graph Index**: A universal graph index that can represent various types of knowledge, including graphs, text, and other structured data. - **Retrieval-Augmented Generation**: The model integrates a large language model with a graph-based retrieval mechanism to enhance its ability to generate relevant and accurate responses. ### Tech Stack: Python ### Why It Matters: The GFM-RAG is a significant advancement in the field of knowledge graph reasoning and language modeling, offering a powerful tool for generating accurate and contextually relevant responses. By combining the strengths of graph neural networks and large language models, it provides a more comprehensive and efficient approach to answering questions and generating text based on structured knowledge.

Overview: The Pixel Reasoner is a model designed to enhance the reasoning capabilities of Large Language Models (LLMs) by enabling them to operate in the pixel-space, specifically through visual reasoning operations. This approach aims to improve the performance of visual tasks by directly inspecting and inferring from visual evidence. Key features: - Supports Instruction Tuning with multi-turn trajectories - Supports RL with multi-turn trajectories - Supports RL with mixed video and image data - Supports Inference/Evaluation with vLLM Tech stack: Python Why it matters: By integrating pixel-space reasoning into VLMs, the Pixel Reasoner addresses the limitation of LLMs confined to textual reasoning, thereby enhancing their performance in visually intensive tasks. This approach has the potential to significantly improve the performance of visual reasoning benchmarks, making it a valuable addition to the field of AI.

Overview: AutoGluon Assistant (aka MLZero) is a multi-agent system designed to automate end-to-end multimodal machine learning or deep learning workflows, transforming raw data into high-quality solutions with zero human intervention. Key features: - Automates end-to-end multimodal machine learning or deep learning workflows - Supports multiple LLM providers, including AWS Bedrock - Provides conversational Q&A mode (`mlzero chat`) for ML guidance without code execution - Offers advanced configuration options and interfaces for customization Tech stack: Python Why it matters: MLZero simplifies complex ML tasks, making it easier for developers to automate and optimize machine learning workflows, ultimately improving efficiency and reducing the need for human intervention.

### Overview: 通过构建企业级 AI 代理系统，从零开始编写每一层，从工具、上下文、架构、编排、可靠性、安全、可观测、评估、并发、成本、发布等方面进行全面训练。项目包含17个双语课程，每课都有带测试的练习，涵盖从零开始到成熟的多进程模块，以及真实生产环境的多进程执行和组件适配。 ### Key features: - **从零开始编写每一层**：从工具、上下文、架构、编排、可靠性、安全、可观测、评估、并发、成本、发布等方面进行全面训练。 - **真实多进程执行**：从第12课起，使用真实的多进程模块，包括崩溃接手、僵尸worker、优雅停机等。 - **成熟组件适配**：Postgres、Redis、Temporal、OpenTelemetry、LiteLLM、Cedar等组件的适配。 - **生产落地**：Postgres、Redis、Temporal、OpenTelemetry、LiteLLM、Cedar等组件的生产落地。 ### Tech stack: - Python 3.10+ ### Why it matters: - 从零开始编写每一层，从工具、上下文、架构、编排、可靠性、安全、可观测、评估、并发、成本、发布等方面进行全面训练。 - 真实多进程执行，包括崩溃接手、僵尸worker、优雅停机等。 - 成熟组件适配，如Postgres、Redis、Temporal、OpenTelemetry、Lite

### Overview: This project is an adapted version of Jina AI's Reader designed for local deployment using Docker. It converts any URL to an LLM-friendly input, making it easy to prepare web content for Large Language Models. ### Key features: - Runs locally using Docker - No API keys required - works out of the box - Saves screenshots locally instead of uploading to Google Cloud Storage - Provides download URLs for saved screenshots - Converts web content to LLM-friendly formats ### Tech stack: TypeScript ### Why it matters: This tool helps streamline the process of preparing web content for Large Language Models, making it easier and more cost-effective.

Overview: This repository contains the code for a paper on explainable prediction of medical codes from clinical text, focusing on multilabel classification of EHR notes using the MIMIC-II and MIMIC-III datasets. Key features: - Multilabel classification - Use of MIMIC-II and MIMIC-III datasets - Pre-trained word embeddings - PyTorch implementation - Jupyter Notebook for data processing - Saved models for reproduction of results Tech stack: Python 3.6, PyTorch 0.3.0, tqdm, scikit-learn 0.19.1, numpy 1.13.3, scipy 0.19.1, pandas 0.20.3, gensim 3.2.0, nltk 3.2.4 Why it matters: The repository provides a framework and code for researchers and developers to understand and improve the explainability of machine learning models in medical text classification, specifically for predicting medical codes from clinical notes.

Overview: SVD-LLM is a method for compressing large language models (LLMs) by using Singular Value Decomposition (SVD) to reduce their size while maintaining performance. This approach is particularly useful for resource-constrained environments. Key features: - Truncation-aware Singular Value Decomposition (SVD) for efficient model compression - Supports LLaMA, Llama-2, and other large language models - Automatically evaluates model performance metrics like perplexity and efficiency Tech stack: Python Why it matters: SVD-LLM provides a practical solution for reducing the size of large language models without significantly compromising their performance, making it valuable for applications where model size is a critical factor, such as resource-limited devices or edge computing environments. Repository: AIoT-MLSys-Lab/SVD-LLM Primary language: Python GitHub stars: 309 GitHub description: [ICLR 2025🔥] SVD-LLM & [NAACL 2025🔥] SVD-LLM V2

### Overview: MoleCode is a molecule representation system that presents molecules as code, enabling large language models to operate and reason on chemistry directly. It converts molecules into a graph format, making topology directly readable, editable, and auditable. ### Key Features: - **Graph Representation**: Molecules are represented as a graph, with atoms as nodes and bonds as edges. - **Persistent Identifiers**: Each atom and bond is assigned a typed declaration with a persistent identifier. - **Inter-Convertibility**: Topology is deterministically and losslessly inter-convertible with SMILES and MOL via RDKit. ### Tech Stack: Python ### Why It Matters: By presenting molecules as code, MoleCode allows large language models to operate and reason on chemistry directly, bypassing the need to reconstruct molecules from positional strings. This approach enables more intuitive and efficient reasoning about molecular structures, reducing the overhead of structural bookkeeping required by traditional methods.

### Overview: SAP Conversational AI is a community platform that allows enterprises and developers to easily create and deploy conversational bots powered by SAP Conversational AI's NLP API. It provides a visual interface for designing conversation flows, webhook features for connecting to external APIs, and a standard messaging format across all channels. ### Key Features: - **NLP API**: A unique API for text processing and augmented training. - **Bot Building Tools**: All you need to create smart bots with rich messaging formats and access to external APIs and services. - **Bot Connector API**: Standardizes messaging formats across all channels, allowing easy integration with any channel in minutes. ### Tech Stack: - Node.js - Python - Java - C++ - JavaScript ### Why It Matters: SAP Conversational AI is a community platform that speeds up bot development by providing a visual interface, webhook features, and standard messaging formats. It also offers a wide range of resources and tutorials to help developers get started and improve their skills. The platform's community nature encourages collaboration and learning from others, making it an excellent choice for developers looking to build conversational bots.

Overview: FreeReg is an image-to-point cloud registration method that leverages pretrained diffusion models and monocular depth estimators to achieve accurate zero-shot registration. Key features: - Unified modality between images and point clouds - Extraction of diffusion features from depth-to-image diffusion models - Extraction of geometric features from monocular depth estimators - Direct utilization of both features for accurate registration Tech stack: Python Why it matters: FreeReg provides a novel approach to image-to-point cloud registration by leveraging pretrained models, which significantly improves accuracy and efficiency compared to existing methods.

Overview: This guide aims to provide an introduction to Retrieval-Augmented Large Language Models (RAGs), focusing on their concept, importance, and key features. The guide will cover the implementation of RAGs, case studies, and applications, aiming to offer a comprehensive understanding of this emerging technology. Key features: - Introduction to Retrieval-Augmented Large Language Models (RAGs) - Explanation of the concept and importance of RAGs - Overview of the key modules and implementation methods of RAGs - Analysis of RAGs in various applications and case studies Tech stack: Python, PyTorch, and other relevant libraries for implementing RAGs. Why it matters: RAGs are becoming increasingly important as they address the limitations of current LLMs, particularly in terms of long-tail knowledge and privacy issues. By combining external databases with LLMs, RAGs can enhance the accuracy and relevance of language and code generation, making them a valuable tool in various applications.

### Overview: BARK is an open-source framework designed for the development, simulation, and benchmarking of behavior planning algorithms in autonomous driving. It is particularly suited for computationally expensive tasks such as reinforcement learning, making it an ideal tool for researchers and developers in the field. ### Key Features: - **Behavior Model Centric Design**: Allows for the rapid development and benchmarking of various decision-making algorithms. - **Python Package**: Available as a pip package for Ubuntu and MacOS, supporting full benchmarking and development of models within Python. - **C++ Build**: Source code available for C++ developers, allowing for the creation of custom behavior models. - **Benchmarking**: Supports the benchmarking of existing behavior models and the development of new models. - **Examples**: Provides examples in both Python and C++ for easy understanding and usage. - **Documentation**: Comprehensive documentation available in both English and Chinese, covering concepts, use cases, and installation instructions. ### Tech Stack: C++ ### Why It Matters: BARK is a valuable tool for researchers and developers in the field of autonomous driving, providing a standardized benchmarking environment for behavior planning algorithms. It simplifies the process of developing and testing new algorithms, making it easier to compare different approaches and improve the overall performance of autonomous vehicles. ### Repository: GitHub Repository: bark-simulator/bark GitHub Stars: 308 Primary Language: C++ ### Why it Matters: BARK is a valuable tool for researchers and developers in the field of autonomous driving, providing a standardized benchmarking environment for behavior planning algorithms

### Overview: This repository is a curated list of machine learning and artificial intelligence startups in Berlin, Germany. It aims to provide developers and enthusiasts with a comprehensive overview of the latest trends and achievements in the field. ### Key Features: - **Natural Language Processing (NLP)** - **deepset**: A platform for developers and enterprise customers. - **explosion.ai**: Tools for machine learning and natural language processing. - **Lateral**: A platform that helps teams with knowledge tasks by capturing expertise. - **lengoo**: A translation service combining AI and linguists. - **LF1**: An AI innovation lab developing proprietary and open-source AI software and technology. - **Voice & Sound** - **Aaron.ai**: Voice processing for companies. - **Audatic**: Sound modification to filter unwanted sounds and enhance interesting ones. - **Pixtunes**: Real-time soundtracks generated by AI. - **Chatbots & Customer Service** - **i2x**: Real-time conversation analytics and coaching for customer interactions. - **RASA**: A framework for building conversational AI applications. ### Tech Stack: Python ### Why It Matters: This list aims to provide a curated and up-to-date overview of the machine learning and AI startups in Berlin, offering developers and enthusiasts a valuable resource for staying informed about the latest trends and achievements in the field.

### Overview: BookWith is an innovative e-book reader that leverages AI to transform reading into a knowledge-creating experience. ### Key Features: - **AI Reading Assistant**: Provides instant answers, deep insights, and context-aware dialogue. - **AI Podcast Generation**: Converts book content into conversational podcasts, optimized for easy listening. - **Multi-Layer Memory System**: Manages short-term, mid-term, and long-term memory to ensure continuity in reading. ### Tech Stack: TypeScript ### Why It Matters: BookWith offers a more engaging and educational reading experience, enhancing comprehension and knowledge retention.

Overview: AubAI is a Flutter/Dart package designed to bring on-device gen-AI capabilities to your apps, including offline text generation, making it a game-changer for developers. Key features: - Text Generation: Harness cutting-edge language models for advanced text generation. - Cross-Platform Support: Compatible with multiple platforms, including macOS, Windows, Linux, Android, iOS, and iPadOS. - User-Friendly Integration: Simplified design for developers of all skill levels. - Open Source: An open-source package, encouraging community contributions and continuous innovation. Tech stack: Dart. Why it matters: AubAI offers developers a powerful tool for integrating on-device gen-AI into their Flutter apps, enhancing user experiences and capabilities without relying on external services.