Projects

Overview: Instant Neural Graphics Primitives is a project that provides implementations of four neural graphics primitives: neural radiance fields (NeRF), signed distance functions (SDFs), neural images, and neural volumes. These are trained and rendered using the tiny-cuda-nn framework with multiresolution hash encoding. Key features: - Neural Radiance Fields (NeRF) - Signed Distance Functions (SDFs) - Neural Images - Neural Volumes Tech stack: Cuda Why it matters: This project offers a fast and efficient way to train and render neural graphics primitives, making it useful for applications requiring real-time rendering capabilities.

### Overview: Discover the latest machine learning / AI courses available on YouTube. ### Key features: - **Machine Learning:** - [Caltech CS156: Learning from Data](#caltech-cs156-learning-from-data) - [Stanford CS229: Machine Learning](#stanford-cs229-machine-learning) - [Making Friends with Machine Learning](#making-friends-with-machine-learning) - [Applied Machine Learning](#applied-machine-learning) - [Introduction to Machine Learning (Tübingen)](#introduction-to-machine-learning-Tübingen) - [Machine Learning Lecture (Stefan Harmeling)](#machine-learning-lecture-stefan-harmeling) - [Statistical Machine Learning (Tübingen)](#statistical-machine-learning-Tübingen) - [Probabilistic Machine Learning](#probabilistic-machine-learning) - [MIT 6.S897: Machine Learning for Healthcare (2019)](#mit-6s897-machine-learning-for-healthcare-2019) - **Deep Learning:** - [Neural Networks: Zero to Hero](#neural-networks-zero-to-hero-by-andrej-karpathy) - [MIT: Deep Learning for Art, Aesthetics, and Creativity](#mit-deep-learning-for-art-aesthetics-and-creativity) - [Stanford CS230: Deep Learning (2018)](#stan

### Overview: ML Visuals is a project aimed at improving scientific communication in the machine learning community by providing free, professional, and compelling visuals and figures. The repository contains over 100 figures contributed by the community, which can be reused and customized for presentations and blog posts. ### Key Features: - **Free and Open Artifacts**: The project provides high-quality visuals and figures that can be downloaded, copied, distributed, reused, and customized. - **Community Contributions**: Figures are contributed by the community, and the repository is updated regularly to include new common figures and basic elements. - **Google Slides**: The visuals are maintained in Google Slides, allowing for easy customization and sharing. ### Tech Stack: - **Google Slides**: Used to maintain and update the visuals and figures. ### Why It Matters: - **Improved Communication**: The project helps machine learning researchers and practitioners enhance their scientific writing by providing visually appealing and informative figures. - **Community Collaboration**: It fosters a collaborative environment where contributors can share their work and ideas, benefiting the entire community. - **Accessibility**: The figures are freely available and reusable, making them accessible to a wide audience. ### GitHub Stars: - **GitHub Stars**: 17,424 ### How to Use: - **Add Custom Figures**: Simply add a new slide and reuse any of the basic visual components. Request edit permissions if necessary. - **Customization**: Provide author information and a short description to help others credit the visuals. - **Download**: Easy to download and use figures from the slides. ### Versions: -

### Overview: turbovec is a vector index built on TurboQuant, a state-of-the-art method for vector search. It is written in Rust with Python bindings, making it accessible to both Rust and Python developers. The library efficiently stores and searches 7.8x smaller embeddings than float32 at 4-bit precision, resulting in 4.4x faster search times compared to FAISS on ARM and x86 architectures. ### Key Features: - **Fast Search**: 4.4x faster than FAISS at 4-bit and 2.2x at 2-bit precision, achieving matching recall at the same speed as FAISS. - **Low Memory**: Stores 7.8x smaller embeddings than float32 at 4-bit and 15.5x smaller at 2-bit, achieving the same recall as FAISS. - **Online**: Adds vectors in microseconds, and removes vectors in milliseconds, making it highly scalable. - **Incremental**: Writes only the changed data, ensuring crash-safe writes at any byte, with a write time of milliseconds even for large indexes. - **Local**: A library, not a service, meaning it can be used in a fully offline environment. - **Filtered**: Allows filtering of results based on a list of allowed IDs or a bitmask, ensuring only relevant results are returned. ### Tech Stack: - **Rust**: The primary language for the library. - **TurboQuant**: The underlying vector search algorithm used for efficient storage and search. ### Why It Matters: turbovec is particularly useful for

### Overview: This repository is a curated library of 462 GPT Image 2 prompts, designed to help developers create, edit, and experiment with images using natural language descriptions. ### Key features: - Browse and copy complete prompts for various image generation and editing tasks. - Adapt placeholders or constraints to your specific needs. - Try the prompts with GPT Image 2 on EvoLink, uploading any input image shown by the case. - Refine variables, composition, style, or output constraints to create the desired image. ### Tech stack: Python ### Why it matters: This library provides a comprehensive set of curated prompts for developers, making it easier to generate and edit images using natural language descriptions.

Overview: This repository contains a curated collection of 462 GPT Image 2 prompts for various applications such as image generation, editing, design, advertising, and visual experimentation. Key features: - Browse prompt categories - Copy complete prompts - Try the prompts with GPT Image 2 on EvoLink - Adapt placeholders or constraints for personal use Tech stack: Python Why it matters: This library provides a comprehensive set of curated prompts for developers to use in their projects, ensuring high fidelity and consistent results.

Overview: Memori is a memory infrastructure designed for enterprise-level systems, seamlessly integrating with existing data infrastructures without requiring a full rearchitecture. It leverages the data and infrastructure already in place, making it ideal for businesses looking to enhance their memory capabilities. Key features: - Agent-native memory infrastructure - LLM-agnostic layer - Persistent state for structured data - Works with existing data infrastructure - Deployed across managed cloud, single-tenant cloud, VPC, and on-premises Tech stack: Python Why it matters: Memori provides a solution for businesses seeking to improve their memory capabilities without the need for a complete rearchitecture of their existing systems. Its agent-native approach ensures that conversations and execution are seamlessly integrated into existing workflows, making it a valuable tool for enterprises looking to enhance their memory infrastructure. Repository: MemoriLabs/Memori Primary language: Python GitHub stars: 17,151 GitHub description: Memori is agent-native memory infrastructure. A LLM-agnostic layer that turns agent execution and conversation into structured, persistent state for production systems. Built for enterprise, Memori works with the data infrastructure you already run, no rip-and-replace, and deploys across managed cloud, single-tenant cloud, VPC, and on-premises. README excerpt: [](https://memorilabs.ai/) <p align="center"> <strong>Memory from what agents do, not just what they say.</

Overview: CVAT is a leading data annotation platform for image, video, audio, and 3D datasets, offering open-source, cloud, and enterprise products, as well as labeling services, AI-assisted labeling, quality assurance, team collaboration, analytics, and developer APIs. Key features: - Open-source, cloud, and enterprise products - AI-assisted labeling, quality assurance, team collaboration, analytics, and developer APIs - Support for image, video, audio, and 3D annotation - Cloud storage integration - Multi-user and multi-organization support - MIT-licensed core - Production-grade foundation for CVAT Online and CVAT Enterprise Tech stack: Python Why it matters: CVAT is a comprehensive tool for data annotation in computer vision, offering a wide range of features and support for various datasets and environments, making it a valuable resource for researchers and developers. Repository: cvat-ai/cvat GitHub stars: 16,945

### Overview: Weaviate is an open-source vector database that combines vector search with structured filtering, fault tolerance, and scalability. It is designed for semantic search at scale, supporting RAG systems, image and semantic search, recommendation engines, chatbots, and content classification. ### Key features: - **Vector Search**: Weaviate allows for efficient vector-based search, enabling semantic similarity and retrieval. - **Structured Filtering**: Combines vector search with structured filtering, providing powerful query capabilities. - **Fault Tolerance and Scalability**: Built-in multi-tenancy, replication, RBAC authorization, and other features ensure robust performance and reliability. ### Tech stack: Go ### Why it matters: Weaviate is a highly scalable and efficient vector database that simplifies complex search and retrieval tasks, making it ideal for applications requiring semantic and image search, recommendation engines, chatbots, and content classification.

### Overview: FinRL® is a financial reinforcement learning framework designed for educational and research purposes. It includes a complete end-to-end solution for financial reinforcement learning. ### Key Features: - **FinRL**: The original framework for financial reinforcement learning. - **FinRL-X**: The latest architecture for AI-native, modular, and production-oriented quantitative trading. ### Tech Stack: Jupyter Notebook ### Why It Matters: FinRL® is widely recognized as the first open-source framework for financial reinforcement learning. FinRL-X offers a next-generation AI-native and production-oriented trading stack. ### Repository: AI4Finance-Foundation/FinRL ### Primary Language: Jupyter Notebook ### GitHub Stars: 16,721 ### GitHub Description: FinRL®: Financial Reinforcement Learning. 🔥

Overview: This repository contains a comprehensive list of PyTorch-related content, including models, implementations, helper libraries, tutorials, and more. Key features: - Pytorch - Captum - NLP & Speech Processing - Computer Vision - Probabilistic/Generative Libraries - Other libraries Tech stack: Python Why it matters: This repository serves as a valuable resource for developers looking to explore and utilize PyTorch and related libraries for various applications in natural language processing, computer vision, and more.

### Overview: This repository is a curated list of Artificial Intelligence resources, designed for software developers looking to build and deploy generative AI and agentic systems. ### Key Features: - **Books**: Comprehensive guides on AI foundations, machine learning, and large language models. - **Courses**: Practical courses on AI engineering, deep learning, and large language models. - **Videos**: Educational videos on AI concepts and practical applications. ### Tech Stack: Python ### Why it Matters: This curated list is essential for developers aiming to understand AI systems, build applications, evaluate AI systems, and deploy them in production. The resources are reviewed and evaluated to ensure quality and practical value, making it a valuable resource for anyone interested in AI development.

### Overview: OpenShell is the safe, private runtime for fleets of autonomous AI agents, ensuring agents can read files, install packages, call APIs, and use credentials without accessing your data, secrets, or network. ### Key features: - **Kernel-level enforcement**: Agents run in isolated sandboxes, with kernel controls confining access to files and system calls. - **Formally verified policy changes**: Before applying policy changes, OpenShell checks for risky new access using formal verification, flagging changes for human review. ### Tech stack: Rust ### Why it matters: OpenShell provides a safe, private runtime for autonomous AI agents, ensuring they can interact with your infrastructure without unauthorized access, enhancing security and compliance.

### Overview: Memvid is a single-file memory layer for AI agents that provides instant retrieval and long-term memory. It replaces complex RAG pipelines with a serverless, memory-centric approach, offering improved accuracy, multi-hop reasoning, and low latency. ### Key Features: - **Instant Retrieval**: Retrieves data quickly and efficiently. - **Long-Term Memory**: Stores and retrieves data persistently and reliably. - **Multi-Hop Reasoning**: Enhances the ability to reason across multiple data points. - **Low Latency**: Achieves ultra-low latency, with a P50 latency of 0.025ms and a P99 latency of 0.075ms. ### Tech Stack: Rust ### Why It Matters: Memvid simplifies AI memory management, making it easier for agents to access and reason with large amounts of data without the need for complex pipelines or databases. This results in faster inference times, improved accuracy, and better performance across various applications. ### Repository: memvid/memvid ### GitHub Stars: 16,602 ### GitHub Description: "Memory layer for AI Agents. Replace complex RAG pipelines with a serverless, single-file memory layer. Give your agents instant retrieval and long-term memory." ### Benchmark Highlights: - **Accuracy**: +35% SOTA on LoCoMo, best-in-class long-horizon conversational recall & reasoning. - **Multi-Hop Reasoning**: +76% multi-hop, +56% temporal vs. industry average. - **Latency**:

Overview: SurfSense is an air-gapped, privacy-focused, open-source alternative to NotebookLM. It is an AI agent that researches, transforms, and edits your documents on your local machine, ensuring your data remains secure. Key features: - Local document research and editing - Support for various document formats - No need for internet connection - User-friendly interface Tech stack: Python Why it matters: SurfSense provides a secure and efficient way to manage your documents without exposing them to the internet, making it ideal for air-gapped environments and ensuring your data remains private and secure. Repository: MODSetter/SurfSense Primary language: Python GitHub stars: 16,505 GitHub description: Air-gapped, privacy-focused, open-source NotebookLM alternative. AI agent that researches, transforms, and edits your documents. Join our Discord: https://discord.gg/ejRNvftDp9

### Overview: ChatALL allows you to concurrently chat with multiple AI bots, including ChatGPT, Bing Chat, Bard, Alpaca, Vicuna, Claude, ChatGLM, MOSS, and more, to discover the best answers. ### Key Features: - **Concurrent Chat**: Send prompts to multiple AI bots simultaneously. - **Discover Best Answers**: Find the most effective responses from various LLMs. - **Supports Multiple Bots**: Includes popular models like ChatGPT, Claude, and Code Llama. ### Tech Stack: JavaScript ### Why It Matters: Whether you're a LLM guru, researcher, or developer, ChatALL helps you quickly find the best-performing models and create optimal prompts.

Overview: Gensim is a Python library designed for topic modeling, document indexing, and similarity retrieval with large corpora. It simplifies the process by providing intuitive interfaces and efficient implementations of popular algorithms. Key features: - Memory-independent processing, capable of handling large inputs even beyond RAM limits. - Easy to plug into custom input data streams and extend with other vector space algorithms. - Efficient multicore implementations for popular algorithms like LSA, LDA, SVD, RP, HDP, and word2vec deep learning. - Distributed computing capabilities for LSA and LDA on clusters. - Extensive documentation and Jupyter Notebook tutorials. Tech stack: Python Why it matters: Gensim simplifies the process of topic modeling, document indexing, and similarity retrieval for large corpora, making it a valuable tool for natural language processing and information retrieval communities. Repository: piskvorky/gensim Primary language: Python GitHub stars: 16,505 GitHub description: Topic Modelling for Humans GitHub release: 16505 Downloads: 16,505 DOI: 10.13140/2.1.2393.1847 Mailing List: Follow @gensim_py on Twitter Sponsor Gensim: Gensim is in stable maintenance mode, focusing on bug and documentation fixes rather than new features.

### Overview Labelme is a Python-based image annotation tool that supports polygon, rectangle, circle, line, point, and AI-assisted annotation. It is designed to be simple and user-friendly, with a focus on providing high-quality annotations for various tasks such as instance segmentation, semantic segmentation, and classification. ### Key Features - **Image Annotation**: Supports polygon, rectangle, circle, line, and point annotations. - **Classification and Cleaning**: Enables image flag annotation for classification and cleaning. - **Video Annotation**: Facilitates video annotation for more complex scenarios. - **GUI Customization**: Offers predefined labels, auto-saving, label validation, and more. - **Exporting Datasets**: Supports exporting VOC-format and COCO-format datasets for semantic and instance segmentation. - **AI-Assisted Annotation**: Integrates with SAM and EfficientSAM models for AI-assisted point-to-polygon/mask annotation. - **AI Text-to-Annotation**: Utilizes YOLO-world and SAM3 models for text-to-annotation. ### Tech Stack - Python - Qt ### Why It Matters Labelme is a powerful tool for image annotation, making it easier and more efficient to create high-quality datasets for various computer vision tasks. Its versatility and user-friendly interface make it an essential resource for developers and researchers alike. ### Repository The Labelme project is hosted on GitHub under the repository name `wkentaro/labelme`. It has received over 16,230 GitHub stars, indicating its widespread adoption and popularity in the community. ### Installation Get started with Labelme

### Overview: Kitten TTS is an open-source text-to-speech library that offers a 1.7B-parameter 1-bit speech language model, allowing users to clone any speaker's voice in real-time on a CPU. It also includes lightweight legacy models for a more efficient alternative. ### Key features: - **Voice Cloning**: Clone any speaker from 5-30 seconds of audio. - **47 Built-in Voices**: Including the eight from KittenTTS 0.8 and nine non-English voices. - **Multilingual Support**: 20 languages: English, Arabic, Chinese, French, German, Hindi, Italian, Portuguese, Russian, Spanish, Japanese, Korean, Turkish, Dutch, Swedish, Danish, Finnish, Swahili, Greek, Hebrew. - **Expression Control**: `[emotion]` tags, inline `<event>` tags, and `(((emphasis)))` spans. - **Decoding Presets**: Trade stability against expressiveness per request. - **Long-form Text**: Sentence-aware chunking with seamless joins. - **24 kHz Output**: High-quality audio at a standard sample rate. - **Runs without a GPU**: Runs real time on a CPU. ### Tech stack: Python ### Why it matters: Kitten TTS is an open-source solution for text-to-speech, offering advanced features like voice cloning and multilingual support, making it a valuable tool for developers and creators looking to enhance their applications with AI-powered speech capabilities.

### Overview: Zvec is a lightweight, in-process vector database designed for fast and scalable similarity search. It is battle-tested within Alibaba Group and offers production-grade performance with minimal setup. ### Key features: - **Blazing Fast**: Searches billions of vectors in milliseconds. - **Simple, Just Works**: Easy installation and straightforward setup. - **DiskANN Productionization**: Adds support for Linux ARM64 and macOS ARM64, as well as an io_uring async I/O backend. - **Index Optimization**: New IVF-RaBitQ index and PQ-INT8 quantizer. - **Deployment Experience Improved**: Slimmed dynamic libraries, support for musl libc and Alpine Linux, prebuilt SDK binaries for various platforms. ### Tech stack: C++ ### Why it matters: Zvec is a lightweight, fast, and efficient vector database that can be easily integrated into applications, providing fast and scalable similarity search capabilities with minimal setup.