Coding & Development

OpenEuroLLM is a project dedicated to creating a series of foundation models for transparent AI in Europe. The initiative focuses on developing strong multilingual models for EU official languages and other socially and economically important languages, ensuring linguistic and cultural diversity. The project emphasizes true openness, including data, documentation, training and testing code, and evaluation metrics, all while complying with EU regulations. OpenEuroLLM aims to provide easy and sustainable access to foundational models ready for fine-tuning across various applications, extend evaluation results and training datasets for European languages, and share transparent tools and processes. It also seeks to build an active community of developers and stakeholders.

Cloudera AI, formerly Verta, is an enterprise-grade platform designed to accelerate the development, deployment, and governance of all types of AI—traditional ML, generative AI, and agentic AI—across hybrid environments. It offers secure, scalable, and governed AI development, ensuring data and model privacy from idea to deployment. Key features include AI Workbench for rapid application building, AI Studios for simplified GenAI development, AI Assistants for enhanced productivity, and the AI Inference service for reliable and scalable model serving. The platform supports low-code to full-code flexibility and allows deployment of models anywhere, with built-in governance and compliance for the entire AI lifecycle.

Vedya Labs provides comprehensive AI solutions designed to accelerate AI adoption and drive innovation across various industries such as semiconductor, tech, robotics, IoT, industrial, HealthTech, and FinTech. The platform offers full-stack AI capabilities, including efficient AI models and integrated systems for physical intelligence and enterprise platforms. Key offerings include Physical AI for real-time, on-device intelligence and Enterprise AI for scalable, autonomous operations with robust data governance. Vedya emphasizes flexible engagement models, leverages proven IP foundations to reduce development cycles, and boasts cross-domain engineering expertise spanning silicon-aware models, embedded systems, and scalable software platforms.

Open Voice OS is a community-driven, open-source voice AI platform designed for creating custom voice-controlled interfaces across a range of devices. It integrates Natural Language Processing (NLP) and provides a customizable user interface, with a strong emphasis on privacy and security. The platform is multi-platform, supporting embedded headless devices, single board computers like Raspberry Pi, and even Linux desktops and laptops. Developers can install it via Docker or Python virtual environments, and pre-built images are available for specific hardware. Open Voice OS allows users to create voice assistants with custom wake words, control smart home devices, play media, get answers, set reminders, and extend functionality through a marketplace of community-developed skills.

2Ai IPCA is a research institute dedicated to the advancement of AI technologies, bringing together experts from diverse fields such as computer vision, machine learning, and natural language processing. The institute's primary goal is to create innovative AI solutions and services applicable across various sectors, including health, industry, environment, and security. As part of the IPCA (Instituto Politécnico do Cávado e do Ave), it operates within a public higher education institution, fostering academic research and development in AI.

REFRAIME offers an advanced AI-powered event detection platform designed to enhance existing surveillance camera systems. By leveraging artificial intelligence, it transforms passive cameras into an active security force, providing real-time monitoring and actionable alerts. The system is capable of instantly detecting events and identifying objects within customized safe zones, delivering alerts via Telegram or web-based platforms. It supports remote monitoring via PC, tablet, and smartphone, and boasts up to a 99.8% reduction in false positive alerts. REFRAIME requires minimal setup, integrating seamlessly with at least one standard surveillance camera and a web-enabled smart device, making it an accessible upgrade for various industries including control rooms, industrial sites, wildlife conservation, residential, commercial, construction, and farming.

QuData provides comprehensive full-stack AI and machine learning solutions, covering everything from initial dataset creation to model design, training, domain knowledge adaptation, and seamless production deployment. The platform offers a wide range of services including LLM integration for chatbots, advanced speech synthesis and recognition, and text analysis for sentiment and content categorization. Additionally, QuData specializes in predictive analytics, computer vision for image analysis, and synthetic data generation for AI training. Their expertise extends to big data processing and medical imaging analysis, leveraging neural networks and deep learning to improve diagnostic accuracy and automate tasks across various industries.

Fay, previously known as PropulsionAI, is an advanced AI agent platform designed to empower non-technical teams to create and manage sophisticated AI agents and workflows. Its core feature, Architect, allows users to simply describe their desired automation in plain language, which then automatically designs, configures, and deploys AI agents. This eliminates the need for coding, configuration files, or engineering expertise. Fay supports the creation of autonomous AI workers that can think, decide, and act towards specific goals, as well as deterministic workflows for reliable automation. It integrates with over 100 tools like Slack, Gmail, and Jira, and supports leading AI models including GPT-4o, Claude, and Gemini, with options to bring your own API keys. The platform emphasizes ease of deployment, security, and enterprise-readiness with features like role-based access control and on-premises deployment options.

Bulbea is a Deep Learning based Python Library designed for stock market prediction and modeling. It provides functionalities for loading historical stock data, preprocessing it for machine learning models, and building recurrent neural networks (RNNs) for price prediction. The library also integrates sentiment analysis capabilities, allowing users to analyze market sentiment from sources like Twitter. Bulbea is built with popular Python libraries such as TensorFlow, Keras, and Tweepy, making it a powerful tool for data scientists and quantitative analysts looking to develop and test algorithmic trading strategies or perform financial forecasting.

skills is an open-source repository by GudaStudio, offering a collection of Agent Skills designed to enhance the capabilities of large language models like Claude. These modular extensions allow Claude to load specialized domain knowledge and workflows on demand, facilitating seamless collaboration with other AI models and tools such as OpenAI Codex and Google Gemini. The project includes skills for delegating coding tasks, enabling prototyping, debugging, and code review. It provides easy installation scripts for Linux/macOS and Windows (PowerShell), supporting user-level, project-level, or custom path installations. The repository also offers recommended global prompts to optimize interaction and state management, asynchronous operations, and a structured workflow for multi-model collaboration, context retrieval, analysis, and code implementation.

Tiny AI is a platform designed to empower users to build their own AI models through an intuitive chat interface. This innovative approach significantly simplifies the often complex process of AI development, making it accessible to a broader audience, including developers and AI enthusiasts. The platform enables users to experiment with artificial intelligence and create custom AI applications without requiring deep technical expertise. By abstracting away much of the underlying complexity, Tiny AI fosters a more creative and experimental environment for AI model building, allowing users to focus on the application and functionality of their AI.

Shannon is an open-source, production-oriented multi-agent orchestration framework designed to help developers ship reliable AI agents to production. It offers robust features such as multi-strategy orchestration, swarm collaboration, and token budget control. The framework includes time-travel debugging for replaying execution steps, real-time event streaming, Prometheus metrics, and OpenTelemetry tracing for visibility. Security is addressed with a WASI sandbox for code execution and OPA policies. Shannon supports a wide range of LLM providers, including OpenAI, Anthropic, Google, and local models via Ollama, with automatic failover. It provides various interaction methods like REST API, Python SDK, and an OpenAI-compatible API, making it versatile for different development workflows.

RAG-Driven-Generative-AI is a comprehensive open-source repository offering programs to construct Retrieval Augmented Generation (RAG) code for Generative AI applications. It integrates powerful frameworks and databases such as LlamaIndex, Deep Lake, and Pinecone, while also utilizing models from OpenAI and Hugging Face for both generation and evaluation. This resource provides a roadmap for building effective LLM, computer vision, and generative AI systems, focusing on balancing performance and costs. It delves into designing, managing, and controlling multimodal AI pipelines, emphasizing how RAG improves output accuracy and contextual relevance by connecting outputs to traceable source documents. The repository includes practical knowledge on vector stores, chunking, indexing, and ranking, along with techniques to optimize project performance and understand data better, including adaptive RAG, human feedback, and dynamic RAG.

handy-ollama is a comprehensive, open-source tutorial designed to help users deploy and manage large language models (LLMs) locally, even on CPU-only systems. This project, officially recognized by Ollama, provides a step-by-step guide from basic installation and configuration across macOS, Windows, Linux, and Docker, to advanced topics like custom model importing (GGUF, Pytorch, Safetensors) and API usage in various programming languages (Python, Java, JavaScript, C++, Golang). It also covers integration with frameworks like LangChain and LlamaIndex, and deployment of visual interfaces using FastAPI and WebUI. The tutorial emphasizes making LLM deployment accessible to everyone, regardless of their computational resources, fostering the development of local AI applications like RAG and Agent systems.

Robovision.gr specializes in delivering cutting-edge machine vision solutions designed to elevate quality control and automation across various industries. Since 2004, Robovision has been a certified provider of automation solutions, developing customized systems that integrate seamlessly into new or existing production lines. Their offerings include 3D surface inspection, color control, hyperspectral imaging, code reading, and deep learning solutions for OCR/OCV. By combining advanced machine vision with robotic guidance, Robovision helps businesses achieve greater accuracy, increased efficiency, and compliance with high industry standards, ultimately boosting productivity and reducing operational expenses.

osgrep is an open-source semantic search tool designed for AI agents, enabling natural language searches within codebases to find concepts rather than just strings. It operates locally and privately, using 100% local embeddings via onnxruntime-node, ensuring data privacy. Key features include call graph tracing to map dependencies, role detection to distinguish orchestration from definition logic, and automatic repository isolation for separate indexing. osgrep is built for speed, offering sub-50ms responses with a lightweight HTTP server and live file watching. It integrates with Claude Code and Opencode plugins, allowing AI agents to query codebases efficiently and save LLM tokens.

FastEmbed is a lightweight and fast Python library designed for generating state-of-the-art embeddings. It supports a variety of popular text and image models, including Flag Embedding for text and CLIP for images. A key differentiator is its use of ONNX Runtime, which eliminates the need for heavy PyTorch dependencies and GPUs, making it suitable for serverless environments like AWS Lambda. FastEmbed offers dense and sparse text embeddings, late interaction models like ColBERT, and multimodal embeddings. It also includes rerankers for improving search relevance. The library is maintained by Qdrant and can be easily integrated with the Qdrant client for vector search applications, with options for GPU support.

RAGLight is a lightweight and modular Python library designed for implementing Retrieval-Augmented Generation (RAG). It significantly enhances the capabilities of Large Language Models (LLMs) by integrating document retrieval with natural language inference. The framework offers modular components, allowing users to easily integrate various LLMs, embeddings, and vector stores to build context-aware AI solutions. A key differentiator is its seamless MCP integration, enabling connection to external tools and data sources. RAGLight supports a wide range of LLM providers including Ollama, Google Gemini, LMStudio, vLLM, OpenAI API, Mistral API, and AWS Bedrock. It also features hybrid search (BM25 + Semantic + RRF), query reformulation, streaming output, and full conversation history support, making it a flexible and powerful tool for developers.

AfterQuery operates as an applied research lab dedicated to advancing foundation model development by curating specialized data solutions. The company addresses the challenge of suboptimal data solutions in AI research by transforming expert knowledge and real-world decision-making into structured training data. AfterQuery's methodology involves capturing how experts think, including their reasoning, decisions, tradeoffs, and context, which is then used to build datasets. Their data offerings include Supervised Fine-Tuning (SFT) with prompt-response pairs and chain-of-thought reasoning, Reinforcement Learning with expert-designed prompts and grading frameworks, Agent Environments for training and evaluating agents in real workflows, and Computer Use Trajectories demonstrating human interactions with software. This approach aims to improve model performance beyond outputs, focusing on enabling models to learn from expert reasoning.

Delfox is an AI-first company that leverages expertise in reinforcement learning and autonomous systems to drive innovation and performance for businesses. They provide end-to-end support, from strategic diagnostics to the implementation of AI solutions. Their services include custom AI development for defense, security, and industry, deployment of AI engineers, and the creation of autonomous decision-making capabilities for drones. Delfox also offers strategic AI consulting to help organizations identify opportunities and build roadmaps. Their proprietary Realmind platform allows users to train, test, and deploy AI models in real-time, utilizing technologies such as generative AI, machine learning, NLP, and computer vision.

OpenRouter offers a unified interface for accessing a wide array of Large Language Models (LLMs) through a single API endpoint. It simplifies the process of integrating various LLMs into applications, providing access to over 300 models from 60+ providers. Key features include higher availability through distributed infrastructure with fallbacks, competitive pricing without sacrificing speed, and custom data policies to ensure prompts are routed to trusted models and providers. The platform is designed for developers, offering an OpenAI-compatible API, client SDKs, and an Agent SDK for building multi-turn agent workflows with tool calls and state management. OpenRouter aims to provide better prices and uptime, operating on a pay-as-you-go model with no subscriptions.

Timespade is an AI development agency based in Hyderabad, India, with offices in San Francisco and Amsterdam, focused on delivering production-grade Minimum Viable Products (MVPs) within 28 days. They offer comprehensive services including generative AI product development (e.g., research copilots, content engines), predictive AI systems (e.g., demand forecasters, fraud scoring), full-stack product engineering for web and mobile applications, and robust data infrastructure solutions. Timespade works with both new ideas and existing products, offering a five-step process from discovery to scaling, with a free initial discovery call. They emphasize fixed pricing per milestone, client ownership of intellectual property, and provide significant partner credits for cloud services and tools to reduce startup costs.

Clade AI provides a comprehensive platform for accessing multiple top AI models, including GPT, Claude, Gemini, Grok, and DeepSeek, all within one application. Users can leverage features like multi-model chat, allowing them to switch between different AI models mid-conversation, and AI Agent tasks for autonomously executing complex multi-step processes. The platform also supports file upload and analysis, web search integration, image generation, and a code interpreter. Available on web and iOS, Clade AI aims to simplify AI access with a single subscription, eliminating the need for multiple accounts and offering a privacy-first approach where user data is never used for AI model training.

RAG-FiT is an open-source library developed by IntelLabs designed to significantly improve Large Language Models' (LLMs) ability to utilize external information within Retrieval-Augmented Generation (RAG) tasks. This framework facilitates fine-tuning models on specially created RAG-augmented datasets. It offers comprehensive support for the entire RAG workflow, including dataset creation, model training using parameter-efficient fine-tuning (PEFT), inference, and performance evaluation with RAG-specific metrics. The library is modular and highly customizable through configuration files, allowing for fast prototyping and experimentation across various RAG settings. It supports integration with external tools and frameworks for information retrieval and prompt generation, making it a versatile solution for developers and researchers working with RAG.