Amazon has launched Nova Forge, a service that allows organizations to build customized AI models tailored to their specific needs. This offering addresses the limitations of existing methods like prompt engineering and Continued Pre-Training (CPT), which often lead to issues like catastrophic forgetting, where models lose foundational knowledge when trained on new data. Nova Forge enables users to blend their proprietary datasets with Amazon’s curated data, starting from various model checkpoints. This approach helps maintain core capabilities while incorporating specialized knowledge. The service is particularly useful for industries with unique data requirements, such as manufacturing, research, and media. Organizations can develop models that understand industry-specific processes, terminologies, and content standards. Nova Forge supports reinforcement learning, allowing models to learn from feedback in environments relevant to the user’s use case. Users can also utilize a built-in toolkit for responsible AI, adjusting safety and moderation settings to fit their business needs. Integration with existing AWS tools, particularly Amazon SageMaker and Amazon Bedrock, makes Nova Forge accessible for users familiar with the AWS ecosystem. The service is currently available in the US East (N. Virginia) region and includes multiple model checkpoints and training recipes to facilitate the development process. Organizations looking for expert support can also connect with Amazon’s Generative AI Innovation Center for assistance in model development.