Projects

Overview: The LangChain Chatbot is an AI chat interface designed to efficiently analyze and extract information from data in conversational format. Key features: - Interactive Ingestion UI for files - Chat UI with adjustable parameters - Compatibility with many more file types and offline models Tech stack: Python Why it matters: This project leverages LangChain for complex tasks, providing a conversational interface for analyzing vector-ingested documents, making it useful for various applications requiring natural language processing and information extraction. Repository: Haste171/langchain-chatbot GitHub stars: 441 GitHub description: AI Chatbot for analyzing/extracting information from data in conversational format.

Overview: A library for ML benchmarking, designed to provide powerful tools for evaluating and comparing machine learning models. Key features: - Supports hyperparameter searches for training models - Contains code for creating LaTeX tables from experiment data - Includes code for evaluating validation methods (validators) - Provides a simple way to run programs via SLURM Tech stack: Jupyter Notebook Why it matters: This library offers a comprehensive suite of tools for benchmarking and evaluating machine learning models, making it a valuable resource for developers and researchers in the field. Repository: KevinMusgrave/powerful-benchmarker GitHub stars: 441 GitHub description: A library for ML benchmarking. It's powerful.

### Overview: A comprehensive collection of OCR and document AI resources, including papers, models, datasets, codes, skills, and platforms, curated and verified by maintainers. ### Key Features: - **Papers**: 331 entries covering research on OCR, document parsing, layout analysis, and document understanding. - **Models**: 883 entries featuring public model cards, weights, APIs, and official model releases. - **Datasets**: 453 entries providing training datasets and evaluation benchmarks. - **Codes**: 154 notable OCR and document AI codebases. - **Skills**: 285 installable agent skills for OCR and document workflows. - **Platforms**: 1 domestic and international OCR platforms and services. ### Tech Stack: Python ### Why it Matters: This collection serves as a valuable resource for developers and researchers interested in automation and document processing, offering a wide range of tools and techniques for OCR and document AI.

### Overview: This project implements the architecture behind the paper "Autoregressive Image Generation without Vector Quantization" by Tianhong Li et al. It provides a PyTorch implementation of the autoregressive diffusion model, enabling efficient image generation without the need for vector quantization. ### Key Features: - **Autoregressive Diffusion Model**: Implements the autoregressive diffusion model architecture. - **Pytorch Implementation**: Provides a complete PyTorch implementation for efficient and flexible use. - **Image Generation**: Supports image generation by treating images as sequences of tokens. - **Alternative Route**: Offers an alternative route via the repository "mar" for those looking to explore the same model. ### Tech Stack: PyTorch ### Why it Matters: This implementation is crucial for researchers and developers looking to explore and implement autoregressive diffusion models, which are essential in generating high-quality images without the limitations of vector quantization. The PyTorch implementation makes it accessible and efficient, facilitating further advancements in image generation and related applications. ### Repository: The repository is hosted at [lucidrains/autoregressive-diffusion-pytorch](https://github.com/lucidrains/autoregressive-diffusion-pytorch). It has been starred by 441 developers, indicating its popularity and utility in the community. ### Install: ```bash $ pip install autoregressive-diffusion-pytorch ``` ### Usage: ```python import torch from autoregressive_diffusion_pytorch import AutoregressiveDiffusion model

Overview: This open-sourced book aims to provide a mathematically light introduction to deep learning, suitable for readers without prior experience in deep learning or strong mathematical background. The book is designed to help readers embark on their journey towards deep learning. Key features: - Chapter 1: Understanding Artificial Neural Networks - Chapter 2: Introduction to PyTorch - Chapter 3: How to make a computer see? - Chapter 5: Where to go from here? Tech stack: CSS Why it matters: This book serves as a valuable resource for beginners in deep learning, offering a gentle introduction to the concepts and tools necessary for the field. Repository: vaibhawvipul/First-steps-towards-Deep-Learning Primary language: CSS GitHub stars: 441

Overview: This repository contains a curated list of essential papers that delve into the planning capabilities of large language models (LLMs). These papers explore various aspects of LLMs' ability to plan, including zero-shot planning, grounded planning, and interactive planning. Key features: - **Zero-Shot Planning**: Extracting actionable knowledge from LLMs for embodied agents. - **Grounded Planning**: Planning for embodied tasks using language models. - **Interactive Planning**: Enabling open-world multi-task agents through LLMs. - **Prompting Techniques**: Improving reasoning and planning in LLMs through various prompting methods. Tech stack: The repository is written in Markdown, HTML, and JavaScript. Why it matters: These papers provide a comprehensive overview of how LLMs can be used to plan and execute tasks, making them invaluable for researchers and developers looking to enhance the capabilities of LLMs in real-world applications.

Overview: SemiKong is an open-source semiconductor AI project that combines a semiconductor language model with an ontology and knowledge graph, providing a comprehensive solution for domain-specific AI workflows. Key features: - A semiconductor language model - A semiconductor ontology and knowledge graph Tech stack: Python Why it matters: SemiKong addresses the need for AI systems that are both fluent and grounded in semiconductor-specific domain structure, crucial for precise technical language, process knowledge, and long operational chains across design, fabrication, test, packaging, and supply. Repository: aitomatic/semikong GitHub stars: 439 GitHub description: First Open-Source Industry-Specific Model for Semiconductors Primary language: Python Why This Project Matters: General-purpose AI is often too shallow for semiconductor work, requiring domain vocabulary, knowledge spanning multiple layers, provenance and validation, and infrastructure for both human and machine understanding. SemiKong aims to fill this gap by providing a comprehensive solution for semiconductor-specific AI workflows.

### Overview: Potato is a free, self-hosted annotation platform designed for NLP, Agentic, GenAI, and qualitative research. It allows annotating text, audio, video, images, documents, agent traces, and more, using YAML configurations. No coding is required, and it supports both live and static annotation interfaces. ### Key Features: - **Text Annotation:** Classification, span labeling, entity linking, coreference, pairwise comparison. - **Agent Traces:** Step-by-step evaluation of LLM agents, tool calls, ReAct chains, and multi-agent systems. - **Web Agents:** Screenshot-based review with SVG overlays, live browsing with automatic trace recording. - **RAG Pipelines:** Retrieval relevance, answer faithfulness, citation accuracy, hallucination detection. - **Audio Annotation:** Waveform visualization, segment labeling, ELAN-style tiered annotation, and 21 transcript/subtitle formats. ### Tech Stack: Python ### Why It Matters: Potato simplifies annotation workflows, making qualitative research and AI evaluation accessible to non-programmers. It supports a wide range of data types and annotation methods, enabling researchers to focus on their research questions rather than the annotation process.

Overview: tinyvector is a tiny embedding database built entirely in Rust, designed for efficiency and ease of customization. Key features: - Extremely easy to customize, with around 600 lines of code. - Should perform comparably to advanced vector databases for small to medium datasets. - Vertically scales well, storing indexes in memory for fast querying. - Open source, licensed under MIT. Tech stack: Rust. Why it matters: tinyvector is a lightweight and efficient solution for embedding databases, making it ideal for simple use cases and easy customization.

Overview: Reverb is an open-source project providing inference and evaluation code for Rev's state-of-the-art speech recognition and diarization models. The project includes ASR and diarization components, utilizing WeNet and Pyannote frameworks respectively. Key features: - ASR: Speech-to-text code based on the WeNet framework. - Diarization: Speaker diarization code based on the Pyannote framework. Tech stack: Python. Why it matters: Reverb provides a platform for developers to access and evaluate advanced speech recognition and diarization models, facilitating the development of applications that can transcribe speech and identify speakers in real-time or as part of a larger system.

### Overview: ViralCutter é um poderoso e gratuito tool para transformar vídeos do YouTube em clips virais para TikTok, Instagram Reels e YouTube Shorts. Com IA de ponta, transcrição dinâmica, face tracking, e edição automática, ele garante resultados profissionais sem custos. ### Key features: - **Corte Viral com IA**: Identifica momentos engajadores automaticamente usando LLMs ou Gemini. - **Transcrição Ultra-Precisa**: Baseada em WhisperX com GPU aceleração. - **Legendas Dinâmicas**: Estilo "Hormozi" com highlight, cores e emojis. - **Direção de Câmera**: Auto-Crop 9:16, Split Screen, Active Speaker. - **Tradução de Vídeo**: Gere legendas traduzidas automaticamente. - **Qualidade & Controle**: Resolução, formatação, e salvações de configurações. - **Performance**: Slicing transcrição, instalação rápida via `uv`. ### Tech stack: Python ### Why it matters: O ViralCutter oferece ilimitado poder no seu hardware, sem necessidade de assinaturas ou limites de minutos. Com 100% privacidade, ele garante que seus dados não saiam do seu PC. Além disso, ele oferece flexibilidade com várias opções de IA e face tracking, além de tradução de vídeos e edição automática.

Overview: This repository contains Keras, PyTorch, and NumPy implementations of various deep learning architectures for Natural Language Processing (NLP). It includes a compact, fully functional, and well-commented PyTorch implementation of the classical seq2seq model "Effective Approaches to Attention-based Neural Machine Translation" by Luong et al. 2015, with support for global attention mechanisms and bidirectional RNNs. Key features: - PyTorch implementation of seq2seq model - Word2vec and doc2vec by hand in NumPy - Hierarchical Attention Network for Document Classification Tech stack: Keras, PyTorch, NumPy Why it matters: This repository provides practical, hands-on examples of deep learning architectures for NLP, making it useful for developers looking to implement and experiment with these models.

Overview: ContactPose is a large dataset of hand-object contact, hand- and object-pose, and 2.9 million RGB-D grasp images. This dataset was introduced in the ECCV 2020 paper "ContactPose: A Dataset of Grasps with Object Contact and Hand Pose." Key features: - Large dataset of hand-object contact, hand- and object-pose, and 2.9 million RGB-D grasp images. - Contains functional grasps with object contact and hand pose. - Useful for training and evaluating grasp prediction models. Tech stack: Jupyter Notebook Why it matters: ContactPose provides a comprehensive dataset for researchers and practitioners working on grasp prediction and related tasks, including hand-object contact, hand- and object-pose, and RGB-D grasp images.

Overview: VLM³ is an official implementation of the paper "VLM³: Vision Language Models Are Native 3D Learners", which shows that standard Vision Language Models (VLMs) are native 3D learners. It surpasses existing state-of-the-art models in object-level 3D understanding, metric depth estimation, and camera pose estimation. Key features: - Surpasses SpatialRGPT on object-level 3D understanding (qualitative and quantitative). - Matches [UnidepthV2] and [Moge-2] on metric depth estimation, improving accuracy from 0.84 to 0.9. - Surpasses [DKM] and [RoMa] for pixel correspondence estimation. - Matches [DepthAnything3] and surpasses [VGGT] for camera pose estimation. Tech stack: Jupyter Notebook Why it matters: VLM³ demonstrates that standard VLMs are sufficient for 3D learning, eliminating the need for complex architectures, losses, and data augmentations. This opens up a new paradigm for simple and scalable 3D learning, making it easier to develop and scale 3D vision models without extensive design efforts. Repository: facebookresearch/VLM3 Primary language: Jupyter Notebook GitHub stars: 438 GitHub description: [NeurIPS 2026] Official implementation of paper "VLM³: Vision Language Models Are Native 3D Learners".

Overview: PyKoSpacing is a Python package designed to automatically space Korean text, improving the accuracy of subsequent text analysis. Key features: - Automatic Korean word spacing - Accuracy for colloquial and literary style corpora - Performance metrics for Sejong and OOOO corpora - Support for common sense validation Tech stack: Python Why it matters: Accurate word spacing is crucial for improving the accuracy of text analysis, especially in online text from sources like social media or SMS. Repository: haven-jeon/PyKoSpacing Primary language: Python GitHub stars: 439 GitHub description: Automatic Korean word spacing with Python

Overview: This is the official supporting code for the book, Grokking AI Algorithms, published by Manning Publications, authored by Rishal Hurbans. The repository contains example implementations of algorithms discussed in the book, intended to provide practical reference for their implementation. Key features: - Python 3.9 or later (3.11 recommended) - PyTorch-compatible GPU for the heavy demos in Chapters 11–12 Tech stack: Python Why it matters: This repository serves as a practical reference for implementing algorithms discussed in the book, providing a technical understanding from a programming perspective.

### Overview: Good-GYM is an AI-powered fitness assistant designed to provide real-time pose estimation, exercise counting, and workout feedback. It uses RTMPose for accurate tracking and supports multiple exercises including squats, push-ups, and dumbbell movements. ### Key features: - Real-time exercise counting - Multiple exercise support - Advanced pose detection - CPU recommended for inference - Visual feedback with skeleton visualization and angle measurement - Workout statistics - User-friendly interface - Works with standard webcams - Runs on Windows, macOS, and iOS ### Tech stack: Good-GYM is built using Python, PyQt5, and RTMPose. ### Why it matters: Good-GYM offers a convenient and accurate way to track your fitness progress, making it a valuable tool for anyone looking to improve their workout routine.

### Overview: PhotonVision is a free, fast, and user-friendly computer vision solution designed for the FIRST Robotics Competition. It offers comprehensive features, including a dedicated Discord community, detailed documentation, and seamless integration with the WPILib toolchain. ### Key Features: - **Documentation**: Accessible through the PhotonVision website, Java docs, and C++ Doxygen. - **Example Projects**: Includes fully featured robot projects and simulation support. - **Community**: Engaging Discord community for contributors. - **Building Instructions**: Detailed guides for compiling and running examples. ### Tech Stack: Java ### Why It Matters: PhotonVision provides essential tools and resources for FIRST Robotics teams, enabling efficient and effective computer vision solutions in the competition. Its user-friendly interface and robust documentation make it an indispensable tool for teams looking to enhance their robot's capabilities.

### Overview: RoboFlamingo is a pre-trained Vision-Language Foundation Models (VLM) framework designed to enable effective robot imitation. It achieves remarkable performance on the CALVIN benchmark, surpassing state-of-the-art results by a large margin. RoboFlamingo can be fine-tuned on a single GPU server, making it a cost-effective and user-friendly solution for robotics manipulation. ### Key Features: - Fine-tuning on offline free-form imitation datasets - Pre-trained models available on Hugging Face - Supports pre-trained vision encoders from OpenCLIP - Supports pre-trained language models from the Transformers package - Cross-attention layers applied every n layers - Decoder types: LSTM, FC, Diffusion, GPT ### Tech Stack: Python ### Why it Matters: RoboFlamingo demonstrates the potential of pre-trained VLMs for robotics control, showing that they can be effectively adapted to robot manipulation tasks. Its performance on the CALVIN benchmark is a significant achievement, indicating its competitiveness in the field of robotic imitation.

### Overview: 这是一个包含超过1000个优秀Qwen提示语的仓库，旨在为Qwen模型提供高质量的提示语示例。这些提示语可以用于生成高质量的回复，增强与Qwen模型的交互体验。 ### Key features: - 多样化的提示语示例，涵盖各种场景和需求。 - 适用于多种语言和文化背景的提示语，确保全球范围内都能使用。 - 提供了如何使用这些提示语的详细指南，帮助用户快速上手。 ### Tech stack: - 仓库托管平台：GitHub - 仓库语言：Unknown ### Why it matters: 这些提示语为Qwen模型提供了丰富的语料库，使得模型能够更好地理解和生成人类语言的回复。此外，通过收集和整理这些提示语，我们希望为开发者和创作者提供灵感和工具，帮助他们创作出更具吸引力的内容。