40 discoveries
aishwaryanr/awesome-generative-ai-guide is a comprehensive, community-driven repository that serves as a central hub for everything related to Generative AI. Created by Aishwarya Naresh Reganti, this project aggregates high-quality research papers, practical notebooks, learning roadmaps, and interview resources. It is designed to help developers, researchers, and AI enthusiasts navigate the rapidly evolving landscape of LLMs, diffusion models, and beyond, making it an essential bookmark for anyone looking to master or stay updated with the latest advancements in the field.
WorldMonitor, developed by koala73, is a sophisticated situational awareness platform that leverages AI to aggregate and analyze global news, geopolitical shifts, and infrastructure status. Built with TypeScript, it provides a unified dashboard for tracking complex events in real-time. By synthesizing vast streams of data into actionable intelligence, the project serves as a powerful tool for analysts, researchers, and organizations needing to maintain a pulse on global developments through an automated, AI-driven lens.
Palmier Pro is an open-source macOS video editor developed by palmier-io, written in Swift. It bridges the gap between traditional non-linear editing and modern AI workflows. By leveraging native macOS frameworks, it provides a performant environment for creators to integrate AI-driven features directly into their video production pipeline, making complex editing tasks more accessible and efficient for desktop users.
LibreTranslate is a powerful, open-source machine translation engine that provides a privacy-focused alternative to commercial translation services. Developed by the LibreTranslate community, it allows users to host their own translation infrastructure. It is built on top of the Argos Translate library, offering a robust API that is easy to deploy via Docker or Python. Because it runs locally, it ensures data privacy and works without an internet connection, making it an ideal solution for developers and organizations needing reliable, sovereign translation capabilities.
MyAppleIntelligence is an open-source project by DaemonLoki that provides a custom implementation of Apple Intelligence-like features. Built primarily in Swift, it leverages CoreML, MLX-Swift, and Llama 3 to bring advanced on-device AI capabilities to Apple hardware. The project serves as a practical exploration of how developers can integrate local large language models and generative AI directly into the Apple ecosystem without relying solely on proprietary cloud-based solutions.
Developed by nealmick, this open-source repository provides a suite of machine learning tools designed for NBA sports betting analysis. It leverages Python, Keras, and TensorFlow to process historical data and generate predictive insights. The project includes infrastructure for data management and model training, enabling users to build custom betting strategies based on statistical probabilities and market odds.
zmre/awesome-security-for-ai is a curated repository maintained by zmre that aggregates essential resources for AI security. It provides a structured list of both open-source and commercial products designed to protect machine learning models and GenAI pipelines. The project serves as a central hub for cybersecurity professionals and AI engineers to discover tools for vulnerability scanning, threat detection, and risk management in AI-driven environments, complete with a helpful infographic for visualizing the AI security landscape.
TradeSight is an open-source, self-hosted AI trading laboratory developed by rmbell09-lang. It provides a comprehensive environment for quantitative finance enthusiasts to design, backtest, and execute trading strategies. The platform features integrated paper trading capabilities, automated overnight strategy tournaments, and support for over 15 technical indicators, allowing users to leverage machine learning to optimize their financial market performance in a controlled, risk-free environment.
Aitino is an open-source platform developed by startino that enables users to orchestrate crews of AI agents. Designed to tackle complex problem-solving and task automation, it provides a framework for multi-agent systems. By leveraging LLMs, Aitino allows developers to define specialized agents that work together to execute intricate workflows, making it a powerful tool for those looking to integrate agentic automation into their projects.
SAP-samples/ai-core-samples is an official repository by SAP providing essential workflow templates and Jupyter notebooks. It serves as a practical guide for developers to integrate, deploy, and manage machine learning models within the SAP AI Core ecosystem. By utilizing the SAP AI Core SDK, users can learn how to transition from experimental code to production-ready AI services, ensuring seamless integration with SAP's enterprise-grade infrastructure.
homemade-gpt-js is an educational project by trekhleb that provides a clean, minimal re-implementation of Andrej Karpathy's minGPT using TensorFlow.js. Designed for developers who want to understand the inner workings of Generative Pre-trained Transformers, the entire model architecture is contained within 300 lines of TypeScript. It serves as a practical, hands-on resource for learning how transformer blocks, self-attention mechanisms, and language modeling work in a browser-based environment.
Forust is a machine learning library developed by jinlow that implements gradient boosted decision trees. Built primarily in Rust for maximum performance, it provides seamless Python integration via PyO3. It serves as a lightweight, efficient alternative to traditional gradient boosting frameworks like XGBoost, focusing on speed and memory efficiency for tabular data tasks.
IBM's Data Science Best Practices is a curated repository designed to bridge the gap between research and production. It provides essential guidelines for data scientists and ML engineers to implement robust versioning, scalability, and monitoring. By focusing on engineering rigor, the project helps teams standardize their workflows, ensuring that machine learning models are not just accurate, but also maintainable, scalable, and reliable in real-world enterprise environments.
bzier/gym-mupen64plus is an open-source OpenAI Gym environment wrapper that integrates the Mupen64Plus N64 emulator with reinforcement learning frameworks. Developed by bzier, this tool allows researchers and developers to train AI agents on classic Nintendo 64 games by providing a standardized interface for state observation and action control. It bridges the gap between complex emulator environments and modern machine learning libraries, facilitating experimentation in game-based AI research.
eren23/one_layer_image_gen is a PyTorch-based repository that replicates the Feature Auto-Encoder (FAE) logic from the research paper 'One Layer Is Enough: Adapting Pretrained Visual Encoders for Image Generation'. It demonstrates how to leverage existing pretrained visual encoders for generative tasks by adding a lightweight, single-layer adaptation, significantly reducing the computational overhead typically associated with training large generative models from scratch.
FoodVisionAI is an intelligent web application developed by Rishabh1925 that utilizes state-of-the-art Vision Transformer (ViT) models to classify food images. By integrating a React frontend with a robust Flask backend, the project provides users with a seamless interface to upload images and receive instant, accurate predictions. It serves as a practical implementation of deep learning for real-world computer vision tasks, demonstrating the effective synergy between Hugging Face model hosting and modern web development stacks.
The MAX78xxx-RefDes repository, maintained by Analog Devices, provides a collection of reference designs for the MAX78000 and MAX78002 microcontrollers. These chips are specifically designed for edge AI, featuring a hardware-based CNN accelerator. The repository includes C-based implementations and examples to help developers integrate deep learning capabilities into power-constrained embedded systems, bridging the gap between high-level machine learning models and low-power hardware execution.
LaMI (Late Multi-Image Fusion) is an innovative framework presented at ACL 2026 that augments Large Language Models (LLMs) with advanced visual capabilities. Developed by Guy Yariv, this approach enables models to process and reason across multiple images effectively. By utilizing a late fusion architecture, LaMI bridges the gap between textual understanding and visual commonsense, allowing for more nuanced multimodal interactions. The official implementation provides the necessary tools for researchers to integrate multi-image context into existing LLM pipelines, pushing the boundaries of vision-and-language tasks.
MjdMahasneh/Simple-PyTorch-Semantic-Segmentation-CNNs is an open-source repository providing clean, modular PyTorch implementations of industry-standard semantic segmentation models. Developed by MjdMahasneh, the project includes architectures such as UNet, DeepLabv3+, SegNet, FCN, and PSPNet. It serves as a practical educational resource and a starting point for developers and researchers looking to implement pixel-level classification tasks in computer vision without the overhead of complex, bloated frameworks.
Low-Rank Decay is an innovative regularization method developed by Chunjiang-Intelligence designed to improve model generalization and combat overfitting. By enforcing low-rank constraints during the training process, this approach helps neural networks achieve better convergence properties. The repository provides the official Python implementation, enabling researchers and developers to integrate this technique into their deep learning pipelines to enhance model performance and stability.
bellerb/chess is an open-source Python program designed for playing chess directly in the terminal. Developed by bellerb, it supports both human-vs-human and human-vs-AI gameplay. The project leverages deep learning and Monte Carlo Tree Search (MCTS) to provide an intelligent computer opponent, making it an excellent resource for developers interested in the intersection of reinforcement learning and classic game theory.
AutismX is an open-source mobile application developed by Mahmoudagha01 using Flutter and Dart. It serves as a dual-purpose tool: providing interactive activities for non-verbal children to improve communication and utilizing AI-driven behavioral analysis to help identify potential autism indicators. The app generates comprehensive reports that parents and caregivers can share with medical specialists to facilitate professional consultations and early intervention.
Developed by the UGR-SAIL research group, this repository provides the supplementary materials, code, and experimental framework for the paper 'An experimental evaluation of Deep Reinforcement Learning algorithms for HVAC control.' It serves as a resource for researchers and engineers looking to apply DRL techniques to optimize building energy consumption and climate control efficiency through intelligent agent-based strategies.
SeeABLE is an open-source deep learning framework designed to identify deepfake content. Developed to address the limitations of traditional binary classification, it utilizes soft discrepancies and bounded contrastive learning to better distinguish between authentic and manipulated media. By focusing on subtle inconsistencies often missed by standard models, SeeABLE provides a more robust approach to forensic analysis in the era of increasingly sophisticated generative AI.
Developed by csinva, this project explores the use of GAN latent space matching to address critical challenges in AI fairness and image manipulation. By identifying and matching latent vectors, the framework allows researchers to perform rigorous bias benchmarking in facial recognition systems and execute fine-grained semantic edits on generated images. It provides a methodological approach to disentangling features, making it a valuable resource for those studying causal inference and fairness in deep generative models.
ApricotM is an open-source framework developed by iheallab for real-time monitoring of ICU patients. Published in Nature Communications, the project leverages state-space modeling (Mamba) and transformer architectures to process electronic health records (EHR). It provides clinicians with predictive insights into patient acuity and therapy requirements, enabling proactive care in high-stakes medical environments by efficiently handling complex, time-series clinical data.
Manifold-linearization is the official companion repository for the research paper 'Representation Learning via Manifold Flattening and Reconstruction.' Authored by Michael Psenka, this project provides the implementation details and Jupyter notebooks necessary to explore how deep learning models can learn representations by flattening complex data manifolds. It serves as a practical resource for researchers and practitioners interested in the intersection of geometry, unsupervised learning, and autoencoder architectures.
Developed by bharatc9530, this project provides a complete pipeline for detecting weapons and fire in images and video streams. Utilizing the YOLOv4-tiny architecture, the repository includes the necessary training code, a curated dataset, and pre-trained weight files. It is designed for integration into surveillance and CCTV systems to enhance security monitoring through automated deep learning-based threat detection.
Developed by HectorPulido, this framework enables seamless integration between Unity and Python for advanced VTuber applications. It leverages deep learning models to perform real-time body tracking and facial expression analysis, allowing developers to drive 3D avatars using AI-powered computer vision. By bridging the gap between Python's robust machine learning ecosystem and Unity's real-time rendering capabilities, it provides a foundation for interactive, motion-captured virtual characters.
Delta is a compact deep learning model developed by GitGud-f, designed for efficient monocular depth estimation. By leveraging knowledge distillation, it compresses the capabilities of the robust DepthAnythingV2 architecture into a smaller, faster model. This approach enables high-quality depth perception on resource-constrained edge hardware, making it an ideal solution for real-time computer vision applications that require low latency and minimal computational overhead.
MyProctorAI is an open-source exam portal developed by hemantkarekar using Flask and Python. It integrates AI-driven proctoring to maintain academic integrity during online assessments. By leveraging image processing and recognition technologies, the system monitors test-takers for suspicious behavior, providing a scalable solution for remote invigilation. It serves as a foundational framework for developers looking to build secure, automated examination environments with integrated anti-cheating measures.
Genet is a specialized Python library developed by Goosang-Yu designed to streamline genome editing workflows. It leverages deep learning models to assist researchers in the design and prediction of CRISPR guide RNAs (gRNAs). By integrating bioinformatics with modern AI techniques, the tool helps scientists optimize gene-editing experiments, improve targeting specificity, and predict the efficiency of CRISPR-Cas systems, making it a valuable asset for genomic research and synthetic biology applications.
ClearML Server for Kubernetes is an open-source deployment solution provided by ClearML that allows teams to host the full ClearML MLOps platform on their own K8s infrastructure. It simplifies the management of experiment tracking, data versioning, and model orchestration by providing standardized Helm charts, ensuring seamless integration with existing Kubernetes-based workflows and private cloud environments.
SM-ViT (Salient Mask-Guided Vision Transformer) is an open-source project by demidovd98, introduced at VISIGRAPP '23. It enhances standard Vision Transformers by integrating a saliency-based masking mechanism. This approach focuses the model's attention on the most discriminative regions of an image, significantly improving performance in fine-grained visual classification tasks where subtle features are critical for accurate identification.
Wildfire-Smoke-Detection is an open-source project by paulinamoskwa that leverages deep learning for environmental monitoring. By utilizing the Faster R-CNN architecture implemented in PyTorch, the model identifies smoke plumes in visual data. This tool provides a foundational framework for researchers and developers aiming to build automated early-warning systems for forest fires, utilizing computer vision to enhance situational awareness and response times in wildfire-prone regions.
Developed by the capstone-insper team, this repository provides a collection of algorithms designed to solve the Drone Swarm Search Environment (DSSE). It leverages deep reinforcement learning and multi-agent systems to optimize how autonomous drone swarms navigate and search for targets. By implementing various strategies, from greedy algorithms to sophisticated deep learning models using PyTorch, the project serves as a practical framework for researchers and developers working on autonomous aerial coordination and critical search-and-rescue applications.
Connectit.chatit is an innovative open-source project by loayabdalslam that reimagines LLM deployment. By utilizing a torrent-like P2P protocol, it enables the decentralized distribution of large language models. This approach reduces reliance on centralized cloud infrastructure, allowing users to share and deploy models across distributed nodes efficiently. It is built with Python and focuses on leveraging peer-to-peer networking to make deep learning models more accessible and resilient in distributed environments.
Risk Distribution Matching is an open-source research project by nktoan, presented at WACV 2024. It addresses the challenge of domain generalization by aligning risk distributions across different training environments. By leveraging causal inference and deep learning, the framework enables models to perform reliably on out-of-distribution (OOD) data, ensuring better generalization in real-world scenarios where test data differs significantly from training data.
Developed by Nevin Baiju, this open-source Android application leverages Google's ML Kit to perform real-time pose estimation and detection. It captures video input to identify human body landmarks and recognizes specific poses performed by the user. Built with Java, the project serves as a practical implementation of on-device machine learning for mobile developers looking to integrate computer vision features into their own Android applications.
This repository contains the source code and methodology used by Kenneth Leung to achieve a Top 5% finish in the Data-Centric AI Competition hosted by Andrew Ng and DeepLearning.AI. Unlike traditional model-centric approaches, this project demonstrates how systematic data cleaning, augmentation, and labeling improvements can significantly boost model performance. It serves as a practical blueprint for practitioners looking to apply data-centric AI principles to real-world machine learning tasks.
