Enhancing Accuracy and Trust in Generative AI with Coveo’s Passage Retrieval API

Co-Authored by Keith Beaudoin and Nicolas Bordeleau from Coveo

In the era of generative AI, how can enterprises ensure that large language models provide accurate and trustworthy responses? Explore how Coveo’s Passage Retrieval API, integrated with Amazon Bedrock, tackles this challenge by delivering context-aware, reliable answers grounded in enterprise knowledge.

Enhancing Trust in AI: The Coveo Passage Retrieval API with Amazon Bedrock

This post is co-written with Keith Beaudoin and Nicolas Bordeleau from Coveo.

As the landscape of business operations evolves with generative AI, organizations confront a vital challenge: ensuring that large language models (LLMs) deliver accurate and trustworthy responses. Without a solid data foundation, the output of AI can veer into the realms of misleading or inaccurate information, jeopardizing user trust and damaging organizational credibility.

The Intersection of AI and Reliable Data

Coveo, as an AWS Partner, is tackling this challenge head-on with its innovative Passage Retrieval API. This powerful solution is designed to enhance the accuracy and trustworthiness of LLM-powered applications by connecting them with relevant, context-aware enterprise knowledge, ensuring that AI-generated responses are grounded in reliable materials.

Central to the process of Retrieval Augmented Generation (RAG) systems, the retrieval component is perhaps the most intricate. It demands extracting the most pertinent and precise information from enterprise data sources. By integrating the Coveo AI-Relevance Platform with Amazon Bedrock Agents, organizations unlock an advanced enterprise search service featuring a secured, unified hybrid index—one that adheres to enterprise permission models and offers robust connectivity.

The Coveo AI-Relevance Platform: A Foundation of Trust

The Coveo AI-Relevance Platform stands out as an industry-leading service that consolidates and connects content from cloud and on-premises repositories into a single index. This unified approach makes it fast and straightforward for users to locate relevant content, improving the experience of retrieving information in complex environments.

Key Features:

1. Relevant Passage Extraction: A two-stage retrieval process identifies the most pertinent documents and extracts the most relevant text passages with accompanying metadata, including source URLs for citations.

2. Hybrid Ranking for Enhanced Results: The combination of semantic (vector) search and lexical (keyword) matching ensures the retrieval of contextually appropriate information.

3. Machine Learning Relevancy: Coveo’s AI learns from user interactions, refining the retrieval process based on individual user journeys and behaviors.

4. Unified Index Across Sources: A centralized hybrid index improves multi-source relevance, outperforming a federated search approach by applying ranking functions across various data sources.

5. Advanced Analytics and Insights: Track performance through the Data Platform and Knowledge Hub, allowing you to identify underutilized content and amplify answer quality.

6. Enterprise-Grade Security: Coveo ensures secure data handling through an early-binding permission model, filtering out restricted content before queries are processed.

Redefining RAG Capabilities

Coveo is reimagining what RAG systems can achieve by transcending basic vector search capabilities. The unified hybrid index seamlessly connects structured, unstructured, and permission-sensitive data, allowing for real-time, ML-driven optimizations. This innovative approach not only enhances response accuracy but also secures organizational data.

Implementing Coveo’s Passage Retrieval API

In this post, we illustrate how to deploy Coveo’s Passage Retrieval API as an Amazon Bedrock Agents action group, empowering Coveo users to leverage their existing knowledge index for rapid deployment of new generative experiences across their enterprises. This functionality is vital for various use cases, from customer support to internal knowledge sharing.

The Integration Process:

1. Set Up the Action Group: Define structured API operations within Amazon Bedrock to enhance user queries with the Passage Retrieval API.

2. Define Agent Instructions: Equip the agent to adeptly retrieve and summarize information regarding Coveo services and capabilities without unnecessary conversational fillers, ensuring concise and informative answers.

3. Establish Backend Connectivity: Utilize AWS Lambda functions to mediate between the Amazon Bedrock agent and the Coveo Passage Retrieval API, facilitating seamless data retrieval based on user input.

Testing and Optimizing Your Solution

Once integration is complete, testing the solution is essential:

• Launch the CloudFormation stack to deploy your agent.
• Enter query examples like, “What is the difference between Coveo Atomic and Headless?” to evaluate the agent’s performance.
• Analyze the logs in Amazon CloudWatch to review retrieved passages and assess their impact on the final answers provided.

Next Steps and Final Thoughts

Organizations that integrate Amazon Bedrock with Coveo’s AI-driven retrieval solution can craft applications capable of delivering validated responses grounded in enterprise content. This intelligent approach diminishes inaccuracies while securing data integrity.

For those looking to dive deeper, explore pre-built examples in our GitHub repository and learn more about implementing the Passage Retrieval API within your Coveo environment through the Passage Retrieval (CPR) implementation overview.

About the Authors

• Yanick Houngbedji: Solutions Architect for AWS, specializing in scalable cloud solutions.
• Keith Beaudoin: Senior Solution Architect at Coveo Labs, expert in intelligent search.
• Nicolas Bordeleau: Head of Product Relations at Coveo, knowledgeable in enterprise and search needs.

By fostering effective use of foundational enterprise data, Coveo and AWS pave the way for a more trustworthy and secure future in the world of generative AI, ensuring that user interactions with AI remain both reliable and credible.