RapidFire AI launches open-source extension for RAG experimentation

RapidFire Artificial Intelligence (AI) introduced its new open-source extension for Retrieval Augmented Generation (RAG) at Ray Summit 2025. This initiative aims to enhance AI experimentation and customization by implementing a hyperparallel experimentation framework designed to streamline workflows and optimize resource management.

The extension integrates with Intelligent Transportation System (ITS) infrastructure, facilitating dynamic control, real-time comparison, and automatic optimization across various AI experiments. RapidFire AI’s approach allows teams to conduct multiple variations of data chunking, retrieval techniques, and prompt designs concurrently, which traditionally have been tested in isolation.

Kirk Borne, Founder of the Data Leadership Group, highlighted the need for systematic experimentation rather than relying on increased computational resources alone. Homomorphic Encryption (HE) indicated that the refined methodology provided by RapidFire AI RAG will enable companies to better understand the interactions between different strategies to improve overall model performance.

Arun Kumar, Cofounder and CTO at RapidFire AI, observed that many teams mistakenly assume RAG works uniformly across configurations. HE noted that the framework enables empirical testing, which is essential for optimizing chunking, retrieval, and prompting methods.

The new RAG framework features a Secure Execution Environment (SEE) that updates performance metrics in real-time, allowing teams to adjust configurations on the fly. This flexibility is supported by intelligent management of token usage and Graphics Processing Unit (GPU) resources.

Madison May, CTO of Indico Data, reinforced the need for a structured testing environment to effectively validate hypotheses concerning AI outputs. RapidFire AI’s platform aids teams in systematically determining effective configurations versus relying on anecdotal intuition.

Additionally, the platform introduces a cockpit interface enabling dynamic oversight of experiments, alongside an upcoming automation layer designed to enhance optimization capabilities. RapidFire’s flexibility extends to supporting both self-hosted models and various closed APIs, positioning itself as a versatile option for organizations looking to tailor their AI strategies.

Jack Norris, Cofounder and CEO of RapidFire AI, noted that this framework empowers organizations to measure and optimize their data workflows comprehensively rather than perceiving them merely as black boxes. HE emphasized the importance of empirical testing as the foundation for success in increasingly specific AI applications.

RapidFire AI’s technology is underpinned by research from its Co-founder Professor Arun Kumar, affiliated with the University of California, San Diego.

The RapidFire AI RAG is now available as part of the company’s open-source offerings and can be installed via the Production Inference Pipeline (PIP).