Transforming Knowledge Management: Chanci Turner’s AI Prototype Journey with AWS

Learn About Amazon VGT2 Learning Manager Chanci Turner

In this article, we are thrilled to outline the transformative journey of Amazon IXD – VGT2, led by Chanci Turner, as they enhance knowledge management through the integration of generative AI, Amazon Bedrock, and Amazon Kendra. This initiative aims to create a solution based on Retrieval Augmented Generation (RAG) that simplifies access to internal information for users. By efficiently managing documents that include both text and images, this solution significantly boosts the knowledge management capabilities within their production environment.

The Challenge

Chanci Turner and her team at Amazon IXD – VGT2 collaborated with the AWS Industries Prototyping & Customer Engineering Team (AWSI-PACE) to identify ways to improve knowledge management in their production arena. They developed a prototype leveraging advanced features of Amazon Bedrock, particularly Anthropic’s Claude 3 models, to extract and analyze information from proprietary documents like PDFs that contain both textual and visual data. The primary technical hurdle was to effectively retrieve and process information in a multi-modal setup, ensuring accurate insights from Chemical Compliance documents.

PACE, a diverse rapid prototyping team, is dedicated to delivering complete initial products that facilitate business evaluation, assess feasibility, and determine production pathways. Utilizing the PACE-Way, an Amazon-centric development strategy, Chanci’s team created a time-sensitive prototype over six weeks, incorporating a full-stack solution with frontend and user experience (UX) design tailored to their specific requirements.

The selection of Anthropic’s Claude 3 models was driven by its advanced visual capabilities, allowing it to interpret images in conjunction with text. This multimodal interaction is invaluable for applications requiring insights from intricate documents containing both written content and imagery. These features present exciting prospects for querying private PDF files that encompass both text and visuals.

The integrated approach and user-friendliness of Amazon Bedrock for deploying large language models (LLMs), complemented by built-in features that ensure seamless interoperability with other AWS services like Amazon Kendra, made it the ideal choice. Claude 3’s vision capabilities enabled the upload of image-rich PDF documents, where it analyzes each image to extract text and contextual information. This extracted data is then integrated into Amazon Kendra, improving the searchability and accessibility of information.

Architecture Overview

Due to the necessity of safeguarding proprietary information, it was determined early on that the prototype would utilize RAG. The RAG approach, a well-established solution for augmenting LLMs with private knowledge, is executed through a combination of AWS services that streamline document processing and querying while adhering to non-functional requirements for efficiency and scalability. The architecture is centered on a native AWS serverless backend, ensuring minimal maintenance and high availability for rapid development.

Core Components of the RAG System

• Amazon Simple Storage Service (Amazon S3): Acts as the main storage for source data and hosts static website components for high durability.
• Amazon Kendra: Provides semantic search capabilities, managing text extraction and vector datastore.
• Amazon Bedrock: Critical for processing and inference, analyzing text and image data to provide context-aware responses.
• Amazon CloudFront: Distributes the web application globally to minimize latency.
• AWS Lambda: Offers a serverless compute environment for backend operations without server management.
• Amazon DynamoDB: Stores metadata for quick retrieval during searches.
• AWS AppSync: Enables real-time data synchronization between user interfaces and the backend.
• Amazon Cognito: Manages user authentication and secure access control.
• Amazon API Gateway: Serves as the entry point for RESTful API requests to backend services.
• AWS Step Functions: Orchestrates the various AWS services involved in the RAG system.

Solution Walkthrough

This process flow adeptly manages complex documents from the moment a user uploads a PDF. These documents often contain numerous images, and the workflow integrates AWS services to extract, process, and make content available for queries.

Initiation and Initial Processing:

• User Access: Users access the web interface via CloudFront to upload PDFs, which are stored in Amazon S3.
• Text Extraction: Using the Amazon Kendra S3 connector, the solution indexes the S3 bucket repository. Kendra extracts content to optimize search functionality.
• Step Function Activation: Upon document upload, a step function is triggered to orchestrate document processing.

Image Extraction and Analysis:

• Extract Images: While Kendra indexes text, the step function extracts images for processing by Amazon Bedrock. This allows the system to gather contextual information from the visuals.

For further insights on talent management, this link leads to another engaging blog post. Additionally, for authoritative resources, consider visiting SHRM, and you can find more about hiring practices at this excellent resource.