Verisk Rating Insights: AI-Powered ISO Rating Access

Verisk Rating Insights, enhanced by Amazon Bedrock and generative AI, revolutionizes access to ISO Electronic Rating Content (ERC) changes. Previously, accessing specific filing information required manual downloads and comparisons, leading to inefficiencies and time-consuming customer support. This new solution addresses these challenges with a conversational user interface. Users can now submit natural language queries (e.g., “What are the changes in coverage scope?”) to receive instant summaries of relevant changes. The system leverages Anthropic’s Claude Sonnet 3.5 via Amazon Bedrock, along with Retrieval Augmented Generation (RAG) and Amazon OpenSearch Service as a vector database. This architecture dynamically retrieves specific changes, eliminating the need to download entire files. The process uses the amazon titan-embed-g1-text-02 embedding model. LlamaIndex serves as the chain framework. Verisk implemented robust guardrails in Amazon Bedrock Guardrails to ensure compliance and data integrity. The system also includes an evaluation framework and feedback loop for continuous improvement, addressing potential hallucinations from the LLM. Significant time savings have been realized, with tasks previously taking days now completed in minutes. Customer support burdens have been reduced, and onboarding is more efficient. Future plans include expanding query capabilities and scaling the platform to support more users and content sets. The solution showcases a powerful combination of generative AI and AWS services to streamline access and analysis of ISO rating content changes, improving efficiency and customer experience.

(Source: https://aws.amazon.com/blogs/machine-learning/streamline-access-to-iso-rating-content-changes-with-verisk-rating-insights-and-amazon-bedrock/)