Projects

Overview: This project provides a Unity SDK for IBM Watson services, allowing developers to integrate Watson-powered applications into their Unity projects. Key features: - IBM Watson SDK for Unity - Tone Analyzer API - Natural Language Classifier API Tech stack: C# Why it matters: This SDK simplifies the integration of Watson services into Unity projects, making it easier for developers to create applications that utilize IBM Watson's AI capabilities.

### Overview: UnifiedSKG:books:: Unifying and Multi-Tasking Structured Knowledge Grounding with Text-to-Text Language Models ### Key features: - UnifiedSKG framework unifies 21 structured knowledge grounding tasks into a text-to-text format. - Promotes systematic and compatible research on structured knowledge grounding. - Achieves state-of-the-art performance on nearly all 21 tasks using large language models like T5. - Facilitates multi-task learning and zero-shot/few-shot learning. - Enables controlled experiments on structured knowledge encoding variants across SKG tasks. ### Tech stack: Python ### Why it matters: UnifiedSKG simplifies the study of structured knowledge grounding by integrating diverse tasks and domains into a unified framework. It facilitates systematic research and promotes compatibility across different communities, making it a valuable resource for researchers and practitioners in the field.

### Overview: Flame is an open-source multimodal AI system designed to translate UI design mockups into high-quality React code. It leverages vision-language modeling, automated data synthesis, and structured training workflows to bridge the gap between design and front-end development. ### Key Features: - Comprehensive Data Preparation Pipeline - End-to-End Training Pipeline - Integration of Siglip Vision encoder and Deepseek coder model for React code generation ### Tech Stack: Python ### Why it Matters: Flame simplifies the process of translating design mockups into functional and scalable React code, addressing the limitations of static code generated by state-of-the-art models. By integrating vision-language modeling, automated data synthesis, and structured training workflows, Flame ensures that generated code is modular, reusable, and dynamic, aligning with best development practices.

Overview: Memory Cache is an experimental project aimed at turning a local desktop environment into an on-device AI agent by saving web pages as PDFs. Key features: - Saves web pages as PDFs - Synchronizes content across multiple devices - Supports privateGPT for augmenting language models Tech stack: JavaScript Why it matters: This project aims to enhance the accessibility and offline capabilities of web browsing, potentially improving the user experience and efficiency of language models. Repository: Mozilla-Ocho/Memory-Cache Primary language: JavaScript GitHub stars: 563 GitHub description: MemoryCache is an experimental development project to turn a local desktop environment into an on-device AI agent

Overview: VSLAM-LAB is a comprehensive framework designed to simplify the development, evaluation, and application of Visual SLAM systems. It provides a unified environment for compiling and configuring VSLAM systems, downloading and processing datasets, and running experiments. Key features: - Unified framework for managing VSLAM systems and datasets - Ease of use with minimal configuration and single command executions - Support for a wide range of VSLAM systems and datasets - Standardized methods for evaluating and analyzing results Tech stack: Python Why it matters: VSLAM-LAB streamlines the development, evaluation, and application of Visual SLAM systems, making it easier to manage, configure, and run experiments. It provides a standardized environment for evaluating and analyzing results, ensuring reproducible outcomes. Repository: VSLAM-LAB/VSLAM-LAB GitHub stars: 563 GitHub description: A Comprehensive Framework for Visual SLAM Systems and Datasets README excerpt: <div align="center"> <img src="docs/header.png" width="500"/> </div> Key features: - Unified framework for managing VSLAM systems and datasets - Ease of use with minimal configuration and single command executions - Support for a wide range of VSLAM systems and datasets - Standardized methods for evaluating and analyzing results Tech stack: Python Why it matters: VSLAM-LAB streamlines the development, evaluation, and application of Visual SLAM systems, making it easier to manage, configure, and run experiments. It provides a standardized environment for evaluating

### Overview: This repository contains the components necessary for applying AllenNLP to a wide variety of NLP tasks, including classification and coreference resolution. It provides easy access to pre-trained models that were trained using these components, making it a valuable resource for developers working on NLP projects. ### Key Features: - **Classification:** Supports tasks like Sentiment Analysis and Binary Question Answering. - **Coreference Resolution:** Detects all expressions in a text that refer to common entities. ### Tech Stack: Python ### Why It Matters: This repository simplifies the process of integrating AllenNLP into NLP projects, making it easier to apply state-of-the-art models and pre-trained models to real-world tasks.

Overview: Skald is a context layer platform designed to provide a production-ready RAG (Retrieval-Augmented Generation) system in minutes. It simplifies the setup and configuration of your RAG engine, making it easy to tailor to your specific needs. Key features: - Skald is a plug-and-play API that handles document parsing, chunking, summarization, tagging, embedding generation, and vector storage. - It supports query rewriting, vector search, LLM chat, chat history, and source references in the retrieval phase. - Skald provides solid defaults but allows for fine-tuning of every part of the RAG to better suit your needs. - Skald can be used with Node.js, Python, Ruby, Go, PHP, C#, and MCP. Tech stack: Skald is built using TypeScript, JavaScript, Python, Ruby, Go, PHP, C#, and MCP. Why it matters: Skald offers a straightforward and efficient way to integrate RAG into your infrastructure, making it accessible and easy to configure. It simplifies the process of setting up and managing a production-ready RAG system, allowing you to focus on your core application without worrying about the complexities of RAG technology. Repository: skaldlabs/skald Primary language: TypeScript GitHub stars: 563 GitHub description: Context layer platform in your infrastructure GitHub link: [skaldlabs/skald](https://github.com/skaldlabs/skald)

### Overview: This repository contains notes and summaries of various papers related to machine learning, computer vision, and natural language processing. The notes are organized by subject, and the summaries are in the form of PDFs. ### Key Features: - **Self-Supervised & Contrastive Learning:** - **"Self-Supervised Relational Reasoning for Representation Learning" (2020)**: [[Paper]](https://arxiv.org/abs/2006.05849) [[Notes]](notes/103_SSL_relation_reasoning.pdf) - **"Big Self-Supervised Models are Strong Semi-Supervised Learners" (2020)**: [[Paper]](https://arxiv.org/abs/2006.10029) [[Notes]](notes/95_big_self-supervised_models.pdf) - **"Debiased Contrastive Learning" (2020)**: [[Paper]](https://arxiv.org/abs/2007.00224) [[Notes]](notes/97_debiased_contrastive_learning.pdf) - **"Selfie: Self-supervised Pretraining for Image Embedding" (2019)**: [[Paper]](https://arxiv.org/abs/1906.02940) [[Notes]](notes/76_selfie_pretraining_for_img_embeddings.pdf) - **"Self-Supervised Representation Learning by Rotation Feature Decoupling

### Overview: This repository is a curated list of projects, libraries, tools, and resources aimed at enhancing and supporting the computational needs of the Arabic language. It includes a wide range of tools for programming languages, such as libraries and applications, as well as resources for translation and text manipulation. ### Key Features: - **Libraries and Tools:** - **libitl2** - A library for Islamic calendar and prayer times. - **Qt** - ArabicNumberToWord, a library for converting numbers to Arabic words. - **goarabic** - A Go Lang package for dealing with Arabic text. - **NumberToArabicWords** - Converts Arabic numbers to words. - **RemoveTashkeel.java** - Removes Arabic diacritics from files. - **Prayer Times** - A desktop application showing prayer times. - **Azkar** - A desktop application for showing Azkar notifications. - **Qafiyah** - An open-source Arabic poetry database. - **I18n** - A repository for providing arabic translations. - **arabicString** - A library for manipulating Arabic strings. - **Arli** - Tools for an Arabic development. - **bidiweb** - Determines the base direction of paragraphs. - **Hijri Converter** - Converts between Hijri and Gregorian calendars. - **miraath-calculator** - Islamic inheritance calculator. - **salah-calculator** - Accurate Islamic prayer times calculator. -

### Overview: AgentSearch is a comprehensive framework designed to power search agents by seamlessly integrating various LLMs with search engines. This framework enables the development of customizable local search engines, enhancing the ability to perform diverse functions through Retrieval-Augmented Generation (RAG). ### Key Features: - **Search Agent Integration**: Seamlessly connect any search-specialized LLM with a supported search engine. - **Customizable Search**: Utilize the AgentSearch dataset to deploy a customizable local search engine. - **API Endpoint Integration**: Seamlessly integrate with a variety of hosted provider APIs, including Bing, SERP API, and AgentSearch, offering flexibility and ease of use. ### Tech Stack: Python ### Why It Matters: AgentSearch simplifies the integration of LLMs with search engines, enabling powerful and customizable search capabilities. This framework is essential for developers looking to enhance their search functionalities and streamline their development processes.

Overview: Obsidian-copilot is a prototype assistant designed to help with writing and thinking, utilizing retrieval-augmented generation techniques. Key features: - Draft sections based on notes - Reflect on the week based on daily journal entries Tech stack: Python Why it matters: It offers a practical solution for enhancing productivity and knowledge retrieval in a structured environment like Obsidian. Repository: eugeneyan/obsidian-copilot GitHub stars: 562

### Overview: The MMDiT repository is an implementation of a single layer of the MMDiT model, proposed by Esser et al. in the paper "Scaling Rectified Flow Transformers for High-Resolution Image Synthesis" (arXiv:2403.03206). This model is designed to scale up the Rectified Flow Transformers for high-resolution image synthesis, addressing the limitations of traditional image synthesis methods. ### Key Features: - **Single Layer Implementation**: Provides a simplified version of the MMDiT model. - **Modality Generalization**: Extends the model to handle multiple modalities (images, audio, and text). - **Adaptive Self-Attention**: Offers an improved variant of self-attention using learned gating, inspired by adaptive convolutions from GigaGAN. ### Tech Stack: - **Programming Language**: Python ### Why It Matters: The MMDiT model is a significant advancement in the field of image synthesis, particularly in high-resolution image generation. By addressing the limitations of traditional image synthesis methods, it opens up new possibilities for creating more realistic and detailed images. The ability to handle multiple modalities makes it a versatile tool for various applications, including augmented reality, virtual reality, and more. ### Repository: lucidrains/mmdit - **GitHub Stars**: 562 - **Primary Language**: Python - **GitHub Description**: Implementation of a single layer of the MMDiT, proposed in Stable Diffusion 3, in Pytorch ### Installation: ```bash $ pip install mmdit

### Overview DoLa is an official implementation of the paper "DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models". It aims to improve the truthfulness of large language models by reducing hallucinations through a simple decoding strategy that contrasts the logits of different layers. ### Key Features - **DoLa**: Decoding by Contrasting Layers - **Truthfulness Improvement**: Enhances the accuracy of LLMs in generating factual content. - **Versatile Application**: Works with various LLM models, including LLaMA family. ### Tech Stack Python ### Why It Matters DoLa provides a practical solution to address the issue of hallucinations in LLMs, making them more reliable in generating factual information. This approach is particularly useful in applications where the accuracy of the generated content is crucial, such as in education, research, and decision-making processes. Repository: voidism/DoLa Primary language: Python GitHub stars: 562

Overview: LangTest is a powerful tool designed to evaluate and enhance language models, ensuring they are safe and effective. Key features: - Generates and executes over 60 distinct types of tests with just one line of code. - Tests multiple aspects of model quality, including robustness, bias, representation, fairness, and accuracy. - Automatically augments training data based on test results for select models. - Supports popular NLP frameworks such as Spark NLP, Hugging Face, and Transformers. - Offers support for testing various types of language models, including question answering, text summarization, and more. Tech stack: Python Why it matters: LangTest is essential for developers and researchers looking to ensure the safety and effectiveness of their language models. By providing comprehensive testing capabilities, it helps identify and mitigate potential issues, ensuring models are reliable and trustworthy. Repository: PacificAI/langtest GitHub stars: 561 GitHub description: Deliver safe & effective language models

Overview: Reaver is a modular deep reinforcement learning framework designed for StarCraft II, supporting Gym, Atari, and MuJoCo environments. It aims to provide both novice and expert researchers with tools to train DRL agents by focusing on StarCraft II tasks. Key features: - Supports StarCraft II, Atari, and MuJoCo environments - Modular architecture with decoupled agent, model, and environment - Focus on StarCraft II tasks, but validated against PPO algorithm results - Easy installation via PIP package or manual installation from source Tech stack: Python Why it matters: Reaver offers a framework for researchers to experiment with deep reinforcement learning in a StarCraft II environment, providing a balance between ease of use for beginners and performance optimization for experts.

### Overview: This repository contains a collection of practical examples showcasing various Data Science and Artificial Intelligence use cases, focusing on machine learning, time series forecasting, natural language processing, and computer vision. Ideal for developers looking to enhance their skills in these areas. ### Key features: - Machine Learning projects including classification, regression, clustering, and recommendation systems. - Time Series forecasting, parametric fitting, and clustering analysis. - Natural Language Processing tasks such as text classification, summarization, and knowledge graph creation. - Computer Vision projects like image classification, object detection, and OCR parsing. ### Tech stack: Python with Jupyter Notebook. ### Why it matters: These examples provide a comprehensive guide for developers to understand and implement Data Science and AI solutions effectively. They cover a wide range of applications, making the content highly relevant and useful for practitioners aiming to improve their skills in these fields.

Overview: ArXiv ChatGuru is a Streamlit app that transforms topics from arXiv into a topic-specific Redis vector index, allowing for efficient retrieval of papers and grounded questions. This project aims to demonstrate how Redis fits into a research Q&A workflow. Key features: - Topic-specific paper chunks and embeddings stored in Redis - Vector search for quick retrieval - OpenAI embeddings for paper chunks - LangChain integration for chat functionality Tech stack: Python Why it matters: This app showcases how Redis can be integrated into a research Q&A workflow, providing a practical example of how semantic search can be applied in academic settings. Repository: redis-developer/ArXivChatGuru GitHub stars: 562 GitHub description: Use ArXiv ChatGuru to talk to research papers. This app uses LangChain, OpenAI, Streamlit, and Redis as a vector database/semantic cache.

### Overview: This repository contains a collection of Jupyter Notebooks designed to accompany the "Building LLMs for Production" book by Towards AI. It serves as a comprehensive resource for understanding and implementing various aspects of using Large Language Models (LLMs) in production environments. ### Key features: - **Transformers Architectures**: Explains the architecture and components of Transformers. - **Understanding Transformers**: Provides an in-depth look at how Transformers work. - **Prompt Engineering Tips and Tricks**: Offers practical advice on crafting effective prompts for LLMs. - **Building Applications Powered by LLMs with LangChain**: Demonstrates how to build applications using LangChain, a library for building LLM-powered applications. - **Building a News Articles Summarizer**: A practical example of using LangChain for summarizing news articles. - **Creating Knowledge Graphs from Textual Data**: Techniques for extracting knowledge from textual data. - **Using Prompt Templates**: Best practices for using prompt templates in LLM applications. - **Managing Outputs with Output Parsers**: Techniques for handling outputs from LLMs. - **Improving Our News Articles Summarizer**: Enhancing the performance of a news article summarizer. - **Creating Knowledge Graphs from Textual Data**: Unveiling hidden connections in textual data. - **Text Splitters and Why They Are Useful**: Understanding the importance of text splitters. - **Chains and Why They Are Used**: Exploring the concept of chains in LLM applications. - **Create a YouTube Video Summarizer Using Whisper and LangChain**: A practical example of

### Overview: 本项目提供了一整套机器学习、深度学习和自然语言处理的基础知识总结，涵盖线性回归、感知机、KNN、朴素贝叶斯、决策树、逻辑回归、最大熵、SVM、AdaBoost、GBDT、EM算法、隐马尔可夫模型、条件随机场、随机森林、XGBoost、聚类、特征工程、神经网络、RNN、LSTM和GRU、CNN、深度学习中的最优化方法、词嵌入（Word2Vec、GloVe、FastText）、TextCNN、Seq2Seq、Transformer、BERT、LSTM+CRF序列标注、主题模型等。 ### Key features: - 线性回归、感知机、KNN、朴素贝叶斯、决策树、逻辑回归、最大熵、SVM、AdaBoost、GBDT、EM算法、隐马尔可夫模型、条件随机场、随机森林、XGBoost、聚类、特征工程、神经网络、RNN、LSTM和GRU、CNN、深度学习中的最优化方法、词嵌入（Word2Vec、GloVe、FastText）、TextCNN、Seq2Seq、Transformer、BERT、LSTM+CRF序列标注、主题模型等。 - 代码示例和笔记文件夹，包含详细的代码实现和笔记解析。 - 代码来自github开源项目，如Seq2Seq、Transformer等部分的代码，部分章节代码待补充。 - 文档文件夹包含笔记、代码

### Overview: Pax is a Jax-based machine learning framework designed for training large-scale models. It offers advanced experimentation and parallelization capabilities, showcasing industry-leading model flop utilization rates. ### Key features: - **Jax-based Machine Learning Framework**: Utilizes Jax for efficient computation. - **Advanced Experimentation**: Supports comprehensive and configurable experimentation. - **Parallelization**: Enables parallel processing for faster model training. - **Industry-Leading Flop Utilization**: Demonstrates superior model flop utilization rates. ### Tech stack: Python ### Why it matters: Pax is crucial for researchers and developers looking to optimize machine learning experiments and accelerate model training on large-scale models. Its advanced features and industry-leading performance make it a valuable tool for researchers in the field of machine learning.