Boost LLM Accuracy with Coveo Passage Retrieval on Amazon Bedrock

Coveo‘s Passage Retrieval API, integrated with Amazon Bedrock Agents, significantly enhances the accuracy of Large Language Models (LLMs) by grounding responses in relevant enterprise data. This solution addresses the challenge of LLMs generating inaccurate or misleading information by providing context-aware knowledge. The Coveo AI-Relevance Platform, a core component, uses machine learning to continuously optimize relevance, unifying cloud and on-premises content sources into a single, secure hybrid index. This unified index allows for a two-stage retrieval process: first identifying relevant documents, then extracting the most pertinent passages with metadata like source URLs. The process incorporates both semantic (vector) and lexical (keyword) search, ensuring accurate information retrieval within the correct context. Machine learning continuously refines retrieval based on user interactions, tailoring results to individual user journeys and profiles. The API offers enterprise-grade security, enforcing permission models from connected sources to prevent data leakage. Integration with Amazon Bedrock Agents is achieved through an action group defined using OpenAPI specifications, allowing the agent to invoke the Coveo Passage Retrieval API via an AWS Lambda function. The solution is easily deployed using AWS CloudFormation, requiring an AWS account, access to Amazon Bedrock models, and a pre-existing Coveo index. Comprehensive analytics and logging capabilities, accessible through CloudWatch, enable performance tracking and iterative improvement. This integration empowers organizations to build more reliable and accurate LLM-powered applications for various use cases, such as virtual assistants and knowledge bots, ultimately improving user trust and organizational credibility.

(Source: https://aws.amazon.com/blogs/machine-learning/enhancing-llm-accuracy-with-coveo-passage-retrieval-on-amazon-bedrock/)