The AWS Generative AI Innovation Center has developed an AI assistant for generating medical content using language learning models (LLMs). Notably, the assistant can reduce content generation time for disease awareness marketing from weeks to hours. Through automation, users can provide the AI with instructions and comments, allowing editing and control over the generation process. The system verifies the accuracy and precision of content, ensuring the produced pieces’ factuality and compliance with set rules and regulations.
The content creation process involves users providing medical references and rules for the marketing content in a brief. The system then uses these inputs in the generation process. The brief lays out the requirements and rules that the generated content should meet, including tone, style, target audience, and word count. Additional rules relating to privacy guidelines and security can also be included.
The UI allows users to upload input data, triggering the Textract functionality, which upon completion, writes the processed data into an S3 bucket. The solution primarily relies on prompt engineering to interact with Bedrock LLMs. It can even guide the LLM further with examples illustrating citation styles. Fine-tuning can further specialize the LLM to medical knowledge— a feature to be explored later.
The system supports multilingual content generation and has provisions to translate content into multiple languages. The revision functionality allows users to improve content over multiple iterations. The AI assistant tool allows users to configure medical content design with a high degree of personalization and seamlessly integrate with existing processes.
The AI assistant ensures the accuracy of generated content, which is essential for highly regulated use cases. It has features to detect any potential misalignment with respect to source references and automatically highlights any violations of rules or regulations. This enables the safe and responsible deployment of AI in industrial settings.
The authors contributing to this topic are Sarah Boufelja Y., Liza Zinovyeva, Nikita Kozodoi, Marion Eigner, Nuno Castro, and Aiham Taleb, who are all part of the AWS Generative AI Innovation Center Team.
