These projects, developers, and companies represent the bedrock for innovation—where open source AI leads the way.
Founded by Australian brothers, Daniel and Michael Han, unsloth’s goal is to make custom AI models more accessible. unsloth fine-tunes open source models 2-5x faster with 70% less memory than its competitors, utilizing emerging techniques and capabilities to make models more performant with maintained accuracy. Diego Marcos, co-creator and maintainer, began developing A-Frame as a framework to make AR/VR and 3D content development accessible to anyone in web browsers. Now focusing on the integration of AI workflows like 3D Gaussian Splatting and generative AI for images and environments, A-Frame has enabled tens of thousands of developers worldwide. The project stands for its commitment to accessibility, community, and availability of resources.
Brazil-based founder, Vinicious Mesel, started working part-time on Talkd.ai to build a unified LLM Chat API that provides an abstraction layer for multiple LLMs and contexts. A first for the Brazilian open source community, the Unified API will enable the LLM to always have and manage context to preprocess the input and generate the prompt from memory or context. Its goal is to facilitate and disseminate the use of the RAG technique in LLMs.UK-based open source advocate Alicia Sykes is a former British Army reservist and previous Oxford intern. Her mission is to make the internet more secure with AI-powered security insights based on open data from any website or server. She built Web-Check to democratize security by making it easier for developers to get a complete view of a website, infrastructure, and server.
Founder, Alex Combessie, CEO, and machine learning R&D engineer, Weixuan XIAO, have built an open source library for testing and evaluating large language models (LLMs). Giskard raises the bar for open source AI model quality, advancing overall adoption, research, transparency, and accountability. Designed for data scientists and developers, Giskard can help ensure the quality, security, and compliance of its customers’ AI models.Founder Namee Oberst is on her second career after a first in law. Together with CEO, Darren Oberst, and Stefan Bachhofner, recognizing the privacy and sensitivity concerns many industries face, Namee sought to build safe and secure LLM AI Agents and Retrieval Augmented Generation (RAG) models for financial and legal institutions. LLMWare provides a comprehensive set of tools that anyone can use—from a beginner to the most sophisticated AI developer—to rapidly build industrial-grade, knowledge-based enterprise LLM applications.
Roboticist Steve Macenski is a pioneer in the Robot Operating System (ROS) navigation framework. Today, Nav2 is used in production worldwide and is the most deployed autonomous mobile robotics (AMR) navigation solution, trusted by more than 100 companies including NVIDIA, Dexory, Polymath Robotics, Stereolabs, and more. Nav2 makes it easy to reliably and efficiently deploy robotics technologies so that users can focus on building their product applications.Cofounders Akshay Agrawal and Myles Scolnick set out to fix all the issues that exist in using notebooks for data science and machine learning. A next-generation Python notebook for AI and machine learning, marimo’s objective is to provide a reproducible, maintainable, and productionizable notebook for AI/ML developers. Today, marimo provides a production-ready notebook that can be deployed as an interactive web app, executed as a script, and versioned with Git.
Founder Tim Baek, based in Canada, wanted to build the best user interface for AI and LLMs to provide opportunities for individuals with limited to no internet access to leverage AI technology and its benefits. OpenWebUI is powered by a web interface that can run LLMs locally making LLMs and AI more secure and private. The project looks to grow both its community of contributors, as well as the project’s reach and impact in communities around the world.Michael Vandi, founder and CEO, and Spatika, founding engineer, built an LLM email agent to respond to emails during a Masters at Carnegie Mellon University (CMU). After some learning, they realized making LLMs accessible with lower GPU needs could benefit more developers and be easier for all. Today, LangDrive serves as a simple framework to train and deploy production-grade fine-tuned language models all via an API and configuration files. This will improve the maintainability of codebases by abstracting the finetuning process and reducing the number of lines for finetuning from hundreds of lines to just 10 lines.
