Projects

### Overview: This repository is a comprehensive list of essential Deep Learning (DL) topics and resources for cracking interviews. It covers a wide range of topics from foundational mathematics to advanced algorithms, providing a solid foundation for aspiring DL practitioners. ### Key features: - **Mathematics**: Includes resources on Linear Algebra, Probability and Statistics, and Information Theory. - **Resources for DL in General**: Lists blogs, books, and open courses for beginners and advanced learners. - **Open Courses**: Provides access to Stanford courses on Machine Learning, Computer Vision, and Natural Language Processing. ### Tech stack: - **GitHub**: 457 stars ### Why it matters: This repository serves as a valuable resource for anyone looking to enhance their knowledge in Deep Learning. It covers essential topics that are crucial for cracking interviews and building a strong foundation in the field.

Overview: DataMind is an all-in-one intelligent assistant powered by LlamaIndex, designed for both personal and enterprise use. It integrates RAG, GraphRAG, NL2SQL, Skills, and Memory functionalities, providing a shared workspace data layer for Codex. Key features: - KB (Knowledge Base): Document chunks and semantic search. - SQL: Exact records, joins, and aggregation. - Graph: Typed entities and relationships. - Skills: Versioned reusable procedures. - Memory: Durable user facts and preferences. Tech stack: Python. Why it matters: DataMind simplifies data management and retrieval, making it easier for Codex to interact with structured and unstructured data, enhancing its reasoning capabilities. Repository: OpenDCAI/DataMind GitHub stars: 455 GitHub description: All-in-one intelligent assistant powered by LlamaIndex — RAG, GraphRAG, NL2SQL, Skills & Memory with multimodal support. Documentation: https://opendcai.github.io/DataMind-Doc/ CI: https://github.com/OpenDCAI/DataMind/actions/workflows/python-ci.yml README excerpt: The shared workspace data layer for Codex. The plugin, official Codex SDK integration, CLI, HTTP API and browser UI share the same data layer. Codex is the main agent for ingestion and retrieval. DataMind contains no StoreAgent/RetrieveAgent implementation or Claude SDK agent backend. Generic native loop primitives remain only for deterministic baseline experiments; product entry points do not assemble them. ## Five surfaces - **KB:** document chunks and semantic search

### Overview: Atticus is a fast, free voice assistant designed for macOS that runs on Claude Pro or Max plans. It seamlessly integrates with your Mac, remembering your shortcuts and learning from your interactions. ### Key Features: - **Instant Commands:** Opens apps, moves windows, plays music, and searches the web without requiring an API key. - **Sees Your Screen:** Answers questions about what's on your screen, including pointing to specific locations. - **Dictate Anything:** Dictates text into any text field, cleaned up. - **Background Agents:** Runs big jobs in the background while you work. - **Meeting Copilot:** Answers and notes during calls, saved on your Mac. - **Remembers You:** Every ask after the initial one remembers your plan. - **No API Key Required:** Runs on Claude, Devin, Pi, or a local model. - **Private and Safe:** Your voice stays on your Mac; asks, sends, and purchases are first. ### Tech Stack: Python ### Why It Matters: Atticus is a powerful tool that enhances productivity without requiring an API key, making it ideal for those looking for a free and efficient solution.

Overview: 这个项目整理了2020-2021年期间计算机视觉领域的综述论文，涵盖了目标检测、图像分割、医学影像、目标跟踪和人脸检测等多个方向。 Key features: - 深度域适应目标检测算法概述 - 平衡问题解决方案综述 - 自动驾驶中的概率目标检测方法综述 - 三维目标检测技术综述 - 伪装目标检测与跟踪研究综述 Tech stack: Python, TensorFlow, PyTorch, OpenCV, Matplotlib等。 Why it matters: 该综述项目为计算机视觉领域的研究人员和开发人员提供了一个全面的视角，帮助他们了解当前研究的最新进展和挑战，从而指导他们的研究和开发工作。

Overview: VRCT is a Python-based chatbox translator and transcription tool designed for seamless integration into VRChat, enhancing communication and accessibility. Key features: - Real-time translation and transcription of chat messages - Integration with VRChat's chat system - Customizable translation settings - Support for multiple languages Tech stack: Python Why it matters: VRCT simplifies communication in VRChat, making it easier for players from different language backgrounds to interact and collaborate, fostering a more inclusive and enjoyable gaming experience.

Overview: This is the official Python SDK for the Pinecone vector database, providing clients for creating, managing, and querying indexes. Key features: - Upsert and query records - Manage indexes - Run inference operations Tech stack: Python Why it matters: The SDK simplifies interaction with Pinecone's vector database, making it easier for developers to integrate vector search capabilities into their applications.

### Overview: This repository contains datasets, evaluation papers, and detection tools for the ChatGPT model. ### Key Features: - **Datasets:** Collection of ChatGPT responses and other related resources. - **Evaluation Papers:** Comprehensive surveys and analyses of various aspects of the ChatGPT model, including NLU, ethics, long text summarization, reasoning, multimodal capabilities, information extraction, machine translation, and more. - **Detection Tools:** Metrics and tools for evaluating the performance of large language models. ### Tech Stack: Unknown ### Why it Matters: This repository serves as a resource for researchers and developers interested in evaluating and understanding the capabilities of the ChatGPT model across various domains and aspects.

### Overview: This repository is a curated list of papers, resources, and collections related to Retrieval-Augmented Generation (RAG) in large language models. It aims to provide a comprehensive overview of the latest research and practical applications in this field. ### Key features: - **Papers**: A collection of surveys and research papers on RAG, including: - A survey of Retrieval-Augmented Generation for Large Language Models (2024) - A survey of Retrieval-Augmented Generation for Natural Language Processing (2024) - A survey of Retrieval-Augmented Text Generation (2022) - A survey of Learning to Retrieve In-Context Examples for Large Language Models (2023) - An active retrieval augmented generation approach (2023) - **Resources**: Includes lectures, talks, and tutorials on RAG, as well as workshops on the topic. - **Tools**: Various tools and libraries used in RAG research and development. - **Other Collections**: Other related collections and resources that may be of interest. ### Tech stack: The primary programming languages used in the repository are Python and JavaScript. ### Why it matters: RAG is a crucial area in the field of large language models, enabling models to generate more contextually relevant and accurate responses by leveraging external knowledge sources. This repository serves as a valuable resource for researchers, developers, and practitioners interested in advancing the state-of-the-art in RAG.

### Overview: Safe & Stable is a user-friendly GUI tool designed to simplify the conversion of Stable Diffusion checkpoints (.ckpt) to the new, secure .safetensors format. This conversion process is crucial for ensuring the safety and stability of your models, especially when dealing with potentially sensitive data. ### Key Features: - **Conversion Between .ckpt and .safetensors**: Easily convert your .ckpt files to .safetensors format and vice versa. - **File Explorer**: A file explorer interface for a more intuitive conversion process. - **Output File Suffix**: Add a suffix to the output file name, keeping the original file name intact. - **Hash Check**: Displays the first 8 characters of the hash for files, ensuring consistency across different formats. ### Tech Stack: - Python ### Why It Matters: This tool is essential for developers and enthusiasts who need to securely store and manage their Stable Diffusion models. By converting your checkpoints to the .safetensors format, you ensure that your models are stored in a more secure and stable manner, reducing the risk of potential security issues.

### Overview: Hayamimi is a real-time multilingual speech-to-text system designed for CPU-only usage. It provides live subtitles, a browser dashboard, speaker labels, and on-the-fly translation without the need for GPU or cloud API. The system routes each utterance to the best-in-class model for the language, achieving a CER of 3.8% on real broadcast Japanese audio, outperforming Whisper variants and cloud STT APIs. ### Key Features: - **5-route language catalog**: Japanese, Chinese, Korean, Yue, and 24 EU languages use dedicated models; others fall back to Meta's Omnilingual ASR. - **Partial subtitles**: Updates in-progress draft text every ~0.5s while you're still speaking. - **Fast finals**: Finalized lines typically land ~100ms after you stop talking. - **Two-pass refinement**: Batch re-decoding after 2s of silence for higher accuracy. - **Speaker labels**: Uses CAM++ nearest-centroid matching and re-diarizes with pyannote segmentation-3.0. - **Translation**: Live translation to English, Chinese, Korean, and Spanish. - **Hotwords / user dictionary**: Biases decoding toward proper nouns and supports post-hoc find/replace. - **CJK number normalization**: Converts kanji-numeral to arabic-digit conversion. - **Runtime dictionary APIs**: Swaps find/replace and ITN exclude/force dictionaries mid-session. - **Structured events + runtime config**: Publishes pipeline stages to an EventHub for app integration. - **O

Overview: Ovis-U1 is a unified model that integrates multimodal understanding, text-to-image generation, and image editing within a single powerful framework, achieving leading scores on academic benchmarks. Key features: - Seamless integration of multimodal understanding, text-to-image generation, and image editing. - A 3-billion-parameter model. - Advanced architecture with a powerful diffusion-based visual decoder (MMDiT) and a bidirectional token refiner. - Synergistic training on diverse data. - State-of-the-art performance across multiple benchmarks. Tech stack: Python Why it matters: Ovis-U1 offers a unified solution for multimodal tasks, improving accuracy and performance across various applications.

Overview: This project is a Mercury application that allows users to train their own custom GPT models by uploading documents or scraping websites. It includes features for chat functionality, embedding support, and a query response system. The project utilizes OpenAI API, Pinecone, Next.js, Tailwind CSS, and Lucide icons for a modern and efficient user interface. Key features: - Train on specific documents or websites - Use Pinecone for vector database storage - Support for various file types (.pdf, .docx, .md, .txt, etc.) - Coming soon: support for .csv, .pptx, and Next 13 app directories Tech stack: TypeScript, OpenAI API, Pinecone, Next.js, Tailwind CSS, Lucide icons, and Radix UI Primitives. Why it matters: This project provides a practical tool for developers and enthusiasts to create custom GPT models, enhancing their ability to automate tasks and improve communication with documents and websites.

### Overview: LightningRAG is a full-stack Vue + Gin starter project that leverages a decoupled frontend and backend architecture. It includes built-in, extensible RAG (retrieval-augmented generation) capabilities, such as knowledge bases, vector search, and integration with various LLMs and vector-store providers. The project is powered by Go and is currently at 454 GitHub stars. ### Key features: - **Knowledge Bases & RAG**: Ingest documents, parse and chunk, embed into pluggable vector stores; hybrid / multi-path retrieval (vector, keyword, PageIndex, and related retriever types); conversation and streaming chat APIs with optional references; pluggable LLMs, embeddings, rerankers, and RAG defaults in `config.yaml` (Section 3 below; [`server/rag/README.md`](server/rag/README.md)). - **Agent Orchestration**: Visual canvas flows with nodes such as Begin, Retrieval, LLM, Message, and Agent (with tools) for multi-step and branching automation; extensible tool registry for RAG chat ([`server/rag/tools/README.md`](server/rag/tools/README.md)). - **Channel Connectors (Optional)**: Bind published agents to webhook endpoints for Feishu, DingTalk, WeChat, Slack, Teams, and other platforms (Section 3.5 below; [`docs/THIRD_PARTY_CHANNEL_CONNECTORS.md`](docs/THIRD_PARTY_CHANNEL_CONNECTORS.md)). - **Authority Management**: Authority management based

Overview: SPSR is a PyTorch implementation of Structure-Preserving Super Resolution with Gradient Guidance, published at CVPR 2020 and TPAMI 2021. This project aims to enhance the quality of low-resolution images by preserving their structural features, making it particularly useful for applications requiring high fidelity reconstructions. Key features: - Structure-Preserving Super Resolution (SPSR) model - Supports both training and testing on various datasets - Preprocess datasets for faster training - Pretrained model for initial parameter initialization Tech stack: Python Why it matters: SPSR offers a significant improvement in the quality of super-resolution images, especially in preserving structural details, making it valuable for applications such as medical imaging, surveillance, and digital restoration.

### Overview: Echogarden is an easy-to-use speech toolset that includes a variety of speech processing tools, written in TypeScript and compatible with Node.js. It supports text-to-speech, speech-to-text, speech-to-transcript alignment, speech-to-translated-transcript alignment, speech-to-transcript-and-translation alignment, text-to-text translation, language detection, voice activity detection, speech denoising, source separation, and word-level timestamps. ### Key Features: - **Text-to-speech**: Supports high-quality models for many languages and dialects. - **Speech-to-text**: Uses Whisper architecture for speech recognition. - **Speech-to-transcript alignment**: Synchronizes speech with an English transcript. - **Speech-to-translated-transcript alignment**: Aligns speech with a translated transcript. - **Speech-to-transcript-and-translation alignment**: Aligns speech with a translation in multiple languages. - **Text-to-text translation**: Translates text between various languages. - **Language detection**: Identifies the language of audio or text. - **Voice activity detection**: Identifies segments of active or inactive voice. - **Speech denoising**: Attenuates background noise from spoken audio. - **Source separation**: Separates voice from music or background ambience. - **Word-level timestamps**: Provides timestamps for all recognition, synthesis, alignment, and translation outputs. - **Enhancements for Kokoro, VITS, and eSpeak-NG**: Improves pronunciation accuracy with text normalization. ### Tech Stack: TypeScript ### Why it Matters: Echogarden is fully open

### Overview: BERT for Coreference Resolution is a state-of-the-art model that improves upon previous methods by leveraging pre-trained language models for coreference resolution tasks. This project includes both a pre-trained model and code for training and evaluating coreference resolution models. ### Key Features: - **BERT-based Coreference Model**: Utilizes pre-trained BERT models to enhance coreference resolution accuracy. - **SpanBERT Model**: Implements a SpanBERT model that improves upon previous methods by representing and predicting spans. - **Pretrained Models**: Includes pre-trained BERT and SpanBERT models fine-tuned on OntoNotes, achieving high F1 scores of 79.6%. ### Tech Stack: Python ### Why it Matters: BERT and SpanBERT models significantly enhance coreference resolution accuracy, making them valuable tools for natural language processing tasks. The pre-trained models provided in this repository offer a practical solution for researchers and developers looking to improve their coreference resolution systems.

### Overview MM-Diffusion is a joint audio-video generation framework that leverages multi-modal U-Net models to produce high-quality realistic videos. It achieves this by learning to generate aligned audio-video pairs from Gaussian noises, resulting in engaging and immersive experiences for both listening and watching. ### Key Features - **MM-Diffusion Framework**: A sequential multi-modal U-Net architecture for generating audio-video pairs. - **Two Subnets**: Audio and video subnets that learn to generate aligned audio-video pairs from Gaussian noises. - **Joint Generation**: Simultaneous generation of audio and video to create a unified and immersive experience. ### Tech Stack - **Python 3.8**: The primary programming language used for implementing the framework. - **PyTorch**: The deep learning framework utilized for building and training the multi-modal U-Net models. ### Why It Matters - **Engaging Experience**: Generates realistic audio-video pairs that provide a seamless and immersive viewing experience. - **High-Quality Output**: Produces high-resolution and realistic video outputs, enhancing the viewing pleasure. - **Versatility**: Can be applied to various datasets, including landscape, AIST++, and custom datasets. ### Repository - **Repository**: [researchmm/MM-Diffusion](https://github.com/researchmm/MM-Diffusion) - **GitHub Stars**: 454 - **GitHub Description**: "This is the official PyTorch implementation of the paper [MM-Diffusion: Learning Multi-Modal Diffusion Models for Joint Audio and Video Generation](https://arxiv.org/

Overview: Simulator Controller is a comprehensive and modular administration tool for sim racing, utilizing the latest GPT technology to provide a realistic and engaging virtual racing experience. Key features: - Comprehensive plugin automation framework for controlling sim racing components. - AI-based race assistants for hands-on assistance during races. - Access to a detailed feature overview in the documentation. Tech stack: AutoHotkey Why it matters: The integration of AI technology in a sim racing application offers a unique and immersive experience, enhancing the realism and enjoyment of virtual racing. Repository: SeriousOldMan/Simulator-Controller GitHub stars: 453 GitHub description: A virtual AI-based pit crew for sim racing, utilizing the latest GPT technology to provide a realistic and engaging virtual racing experience.

### Overview: AVA (Aesthetic Visual Analysis) is a large-scale database designed for aesthetic visual analysis, containing 250,000+ photos from dpchallenge.com. This dataset is crucial for researchers and developers interested in studying visual aesthetics and human perception. ### Key Features: - **Dataset Size**: 250,000+ photos. - **Format**: 64 7z files. - **Download Method**: Available via Kaggle, HuggingFace, Mega Cloud, and torrent. ### Tech Stack: - **7z File Format**: The dataset is stored in 7z files. ### Why It Matters: AVA provides a comprehensive resource for researchers and developers to study visual aesthetics, human perception, and machine learning techniques applied to visual analysis. It is essential for anyone interested in advancing the field of visual analysis and artificial intelligence in the domain of aesthetics. ### Repository: - **Repository**: imfing/ava_downloader - **GitHub Stars**: 453 - **GitHub Description**: :arrow_double_down: Download AVA dataset (A Large-Scale Database for Aesthetic Visual Analysis) ### Reference: - **Bibtex**: @INPROCEEDINGS{6247954, author={Murray, Naila and Marchesotti, Luca and Perronnin, Florent}, booktitle={2012 IEEE Conference on Computer Vision and Pattern Recognition}, title={AVA: A large-scale database for aesthetic visual analysis}, year={2012}, volume={}, number={}, pages={240

Overview: MLX-Embeddings is a package designed to run Vision and Language Embedding models locally on your Mac using MLX, making it accessible for developers and researchers. Key features: - Generate embeddings for text and images using MLX models - Support for single-item and batch processing - Utilities for comparing text similarities Tech stack: Python Why it matters: MLX-Embeddings simplifies the process of running Vision and Language Embedding models locally, making it easier for developers to experiment with and integrate these models into their projects. Repository: Blaizzy/mlx-embeddings Primary language: Python GitHub stars: 449