Coding & Development

X-modaler is an open-source, high-performance codebase designed for cross-modal analytics, encompassing a wide range of tasks such as image captioning, video captioning, vision-language pre-training, visual question answering, visual commonsense reasoning, and cross-modal retrieval. It offers a unified collection of high-quality modules for state-of-the-art vision-language techniques, organized in a standardized and user-friendly manner. The codebase supports various models including LSTM-A3, Up-Down, Transformer, and TDEN across different tasks, providing baseline results and trained models for research and development. It requires Python 3.6+, PyTorch 1.8+, and other specific libraries, making it suitable for technical users and researchers in AI and machine learning.

what_are_embeddings is an open-source GitHub repository dedicated to exploring the fundamentals, history, and industrial usage patterns of embeddings in machine learning. The project includes a comprehensive LaTeX document, a generated website, and supporting notebook code, making it a valuable resource for anyone looking to understand these numerical representations. It covers the evolution from traditional methods like TF-IDF and PCA to modern approaches enabled by Word2Vec and Transformer architectures. The repository is designed for educational purposes, offering a deep dive into how embeddings scale with increasing data volume, velocity, and variety in modern applications. Users can contribute to the document by building the LaTeX artifact and submitting pull requests.

voxtral-mini-realtime-rs is an open-source project offering real-time streaming speech recognition (ASR) and text-to-speech (TTS) functionalities. Built with Rust and leveraging the Burn ML framework, it implements Mistral's Voxtral Mini 4B Realtime ASR and Voxtral 4B TTS models. The tool is designed for both native execution and in-browser use via WASM + WebGPU, making it highly versatile. It supports Q4 GGUF models for efficient, client-side operation in a browser tab, addressing challenges like allocation limits and GPU readback. Key features include 20 preset voices across 9 languages for TTS, and optimizations like batched CFG and pre-allocated KV cache for ASR. Benchmarks demonstrate its performance for both ASR and TTS, with options for BF16 and Q4 GGUF models.

vqa.pytorch is an open-source project offering a PyTorch implementation for Visual Question Answering (VQA). Developed by researchers at LIP6 and Heuritech, this tool aims to facilitate the reproduction of state-of-the-art results, particularly those achieved with the MUTAN: Multimodal Tucker Fusion for VQA method on the VQA 1.0 dataset. It provides a modular and efficient codebase for further research on various VQA datasets. Key features include support for different VQA datasets (VQA 1.0, VQA 2.0, VisualGenome), pretrained models, and tools for extracting features from images using convolutional neural networks. The repository also includes documentation on its architecture, options, and quick examples for training and evaluating models, making it a valuable resource for researchers and students in the field of computer vision and natural language processing.

Yet Another Applied LLM Benchmark is an open-source tool designed to evaluate the performance of large language models (LLMs) on practical, real-world tasks. The benchmark is unique because its tests are derived directly from questions the creator has previously asked LLMs to solve, covering scenarios like converting Python to C, decompiling bytecode, explaining minified JavaScript, and generating SQL queries. It features a simple dataflow domain-specific language (DSL) that allows users to easily create new, sophisticated test cases. This DSL enables complex evaluation flows, such as asking an LLM to generate code, running that code in a Docker container, and then using another LLM to evaluate its output. The tool emphasizes testing models on tasks that developers genuinely care about, providing a more realistic assessment than many academic benchmarks.

Jait is an innovative local-first AI workstation designed to streamline the development and interaction with AI agents. This powerful platform unifies essential developer tools, including chat, terminals, file editing, browser control, screen sharing, and automation threads, into a single, cohesive environment. It caters to developers, AI engineers, and researchers who require a robust workspace for complex AI projects. Jait supports a wide array of AI models, allowing users to integrate popular services like OpenAI Codex, Claude Code, and Gemini CLI, or even bring their own local models via Ollama and LM Studio. By keeping the working context visible and integrating source control directly into the conversation, Jait enhances productivity and fosters a more intuitive development workflow for building and managing intelligent agents.

Comma AI offers an advanced driver-assistance system powered by openpilot software, designed to make driving more relaxed. The system provides key features such as lane centering, adaptive cruise control, and dashcam recording, enhancing the driving experience. It also includes lane changing capabilities and 360° vision. The comma four hardware integrates seamlessly with a wide range of vehicles, supporting over 325 car models from 27 brands, including Toyota, Hyundai, and Ford. Users can easily purchase the device, plug it in, and engage the system. Software updates are delivered over-the-air (OTA). Comma AI emphasizes driver alertness with a camera-based Driver Monitoring (DM) system and ensures safety by allowing immediate manual control. The platform also features a community-driven approach with a strong GitHub presence and a support system for hardware and software issues.

ReBatch, an AI company within the Cronos Group, specializes in delivering tailor-made AI solutions with a strong emphasis on data sovereignty and EU compliance. They offer three deployment models: Sovereign AI for on-premise infrastructure with maximum data control, Hybrid AI combining on-premise security with cloud scalability, and Cloud AI utilizing EU-hosted infrastructure for data residency. ReBatch builds intelligent assistants, agentic workflows, and knowledge platforms, ensuring full customer control. Their expertise covers Agentic AI, AI in SDLC, Knowledge Assistants via their iriSearch platform, and Document AI, all designed to meet strict regulatory requirements like GDPR, EU AI Act, and NIS2.

DefSafe is an AI-powered cybersecurity platform designed to protect organizations against cyber threats. It offers comprehensive cybersecurity awareness training with over 90 AI-guided modules, covering topics from basic security awareness to advanced concepts like OWASP and emerging technologies. The platform also includes live vulnerability scanners, called AI Sentinel, to detect weaknesses in websites and provide comprehensive fixes. DefSafe emphasizes a defense-in-depth strategy, offering role-specific training and customizable programs with features like AI-guided learning, accountability tracking, and shareable certificates. It aims to empower teams with the knowledge and tools needed for robust cyber compliance and safety.

The Open Medical-LLM Leaderboard is a platform dedicated to evaluating open large language models specifically designed for medical applications. Users can browse and filter a comprehensive leaderboard of various models, comparing their performance across different medical-related tasks. The platform also enables users to submit their own models for evaluation, fostering a collaborative environment for advancing medical AI. It is a valuable resource for researchers and healthcare professionals looking to assess and select the most suitable LLMs for their specific needs in the medical domain, offering insights into model precision, type, and size.

one-pixel-attack-keras is an open-source project offering a Keras implementation of the "One pixel attack for fooling deep neural networks." This tool demonstrates how minimal perturbations, specifically changing just one pixel's color, can cause deep neural networks to misclassify images. It leverages differential evolution on datasets like Cifar10 and ImageNet to iteratively generate adversarial images and minimize the network's classification confidence. The project includes tutorial notebooks, various CNN models (LeNet, ResNet, DenseNet, CapsNet), and scripts for training and attacking models. It's particularly useful for understanding and researching the robustness and vulnerabilities of deep learning models to adversarial attacks.

Stable Horde is a community-powered network where volunteers contribute their computing resources to facilitate free AI image and text generation. Users can access the platform through various frontends, such as ArtBot, to visually create content without needing coding or complex setup. The platform operates on a 'Kudos' system, where workers earn Kudos for processing jobs, and users spend Kudos for higher priority in the generation queue. Kudos never expire and represent a user's contribution to the community. Stable Horde emphasizes accessibility, offering anonymous usage with a basic API key, though unique API keys and accumulated Kudos provide better priority. It aims to provide a free and collaborative environment for AI content creation.

anago is a Python library designed for sequence labeling tasks, including Named Entity Recognition (NER) and Part-of-Speech (PoS) Tagging. Built with Keras, it leverages advanced models like Bidirectional LSTM-CRF and ELMo to achieve high performance. A key differentiator is its independence from language-dependent features, making it easily adaptable for various languages. The library offers essential methods for model training, evaluation, and text tagging, along with support for custom models, pre-trained model downloads, and GPU acceleration. It's particularly useful for researchers and developers working on natural language processing applications.

SharpAI Aegis transforms your Mac or PC into a powerful home AI security agent, integrating seamlessly with Ring, Blink, and RTSP/ONVIF cameras. It downloads clips and feeds them to local LLM/VLM models for advanced video analysis, recognizing family members, spotting strangers, and identifying objects like packages. The system runs entirely on your hardware, ensuring privacy and eliminating cloud subscriptions. Users can ask natural language questions about events, receiving timestamped answers and summaries instead of endless motion alerts. Aegis provides a unified timeline for all camera events, with AI automatically analyzing clips and highlighting important detections. It supports a wide range of cameras and offers both local AI processing for privacy and optional cloud model integration for enhanced capabilities.

Terracotta AI is an AI-powered infrastructure intelligence platform designed for robust change governance across Terraform, Kubernetes, and Terragrunt. It automates 12 critical checks on every pull request, including cost estimation with live AWS and GCP pricing, drift detection across 119 AWS resource types, and comprehensive security and IAM analysis. The platform also enforces guardrails in plain English, performs blast radius analysis, and offers architecture visualization. With an AI code review feature and a fleet-wide command center dashboard, Terracotta AI ensures infrastructure changes are secure, compliant, and cost-effective before they are merged, streamlining development workflows for technical teams.

Butterfly GAN is an AI image generator specifically designed for creating butterfly images. This tool operates as a Hugging Face Space application, leveraging the Streamlit framework for its user interface. It is licensed under Apache-2.0, making it suitable for various uses, including educational exploration of generative adversarial networks (GANs). While the current live website indicates a runtime error, the tool's core purpose is to demonstrate AI's capability in generating specific image types, offering a focused approach to image creation within the butterfly domain.

auto-diffuser-config is an application designed to assist users in generating optimized code for image generation tasks. It simplifies the process by allowing users to input their hardware details and desired model settings. The tool aims to provide detailed configurations, making it easier for developers to set up their AI models efficiently. While the current status indicates a runtime error, its intended purpose is to streamline the code generation process for AI applications, particularly those utilizing the Diffusers library, by tailoring code based on specific hardware and model requirements.

ArXiv New ML Datasets is a specialized tool designed to help researchers and academics discover new machine learning datasets within the vast collection of arXiv computer science papers. Users can efficiently search for relevant papers using either keyword-based queries or advanced semantic search capabilities. The platform then allows for further refinement of results by research category, making it easier to pinpoint specific areas of interest. This tool is particularly valuable for those looking to stay updated on the latest dataset introductions in the machine learning field, facilitating academic research and data-driven projects by providing a focused and streamlined discovery process.

Awesome-Visual-Transformer is a comprehensive, open-source repository dedicated to collecting and organizing academic papers focused on the application of transformers in computer vision (CV). This tool serves as an invaluable resource for researchers, academics, and practitioners looking to stay updated on the latest advancements in this rapidly evolving field. The collection includes original transformer papers, surveys, and numerous arXiv preprints covering diverse topics such as 3D semantic segmentation, object detection, image generation, medical image synthesis, and video processing. Users can easily browse papers, often with links to associated code, making it a practical resource for both theoretical understanding and implementation. The repository encourages community contributions through issues and pull requests, fostering a collaborative environment for knowledge sharing.

auto-gpt-web is an open-source AI agent that empowers users to define initial roles and goals for an AI buddy, which then operates autonomously to achieve these objectives. Inspired by Auto-GPT, this tool features internet access for comprehensive searches and information gathering. A key differentiator is its local storage capability, saving AI definitions, chat history, and credentials directly within the user's browser, ensuring privacy and control. It also includes long-term memory based on a browser-based vector database and an Electron application for conducting search operations, bypassing typical API limitations. Users need an OpenAI API Key and Google Search API Key with a Custom Search Engine ID to utilize its full capabilities.

BLOOMChat is an accessible and free-to-use multilingual chatbot model, hosted on Hugging Face Spaces. It is built upon the BLOOM (176B) model and has been instruction-tuned for assistant-style conversations. Users can engage with the AI to get information, ask questions, or simply have a conversation in various languages. The platform emphasizes ease of access, requiring no sign-up or personal information, making it a straightforward tool for quick interactions and explorations of conversational AI capabilities. Its open nature on Hugging Face Spaces also suggests a community-oriented approach to AI development and accessibility.

awesome-llm-security is a curated GitHub repository dedicated to providing a comprehensive collection of resources focused on Large Language Model (LLM) security. It serves as a valuable hub for security researchers, AI developers, and cybersecurity professionals seeking to understand and address vulnerabilities in LLMs. The repository categorizes resources into white-box, black-box, and backdoor attacks, as well as fingerprinting, defense mechanisms, platform security, surveys, and benchmarks. It also lists various tools like Plexiglass, Rebuff, Garak, LLMFuzzer, and LLM Guard, alongside relevant articles and other awesome projects. This resource aims to foster better security practices and research within the LLM ecosystem.

CL EVA02 LoRA ONNX Tagger is an AI tool designed for image tagging, specifically for anime images and illustrations. Users can upload an image or provide an image URL to receive predicted tags that describe its content. The tags are categorized into types such as rating, general, and character. The tool also offers a visualization of the generated tags, providing a comprehensive overview of the image's characteristics. It utilizes ONNX models for efficient image classification, making it suitable for tasks like organizing image datasets and supporting computer vision research.

This repository offers a PyTorch implementation of the paper "Evidential Deep Learning to Quantify Classification Uncertainty." It provides an accessible and easy-to-run demonstration of the concepts presented in the paper, requiring low computational resources. The tool allows users to explore how neural networks can be trained to quantify their prediction confidence, moving beyond standard softmax probabilities. It includes implementations for various loss functions, such as Expected Mean Square Error, Expected Cross Entropy, and Negative Log of the Expected Likelihood, enabling different approaches to uncertainty estimation. The project highlights how this method improves uncertainty detection for out-of-distribution queries and enhances resilience against adversarial perturbations, making it valuable for researchers and developers working on robust AI models.