Qalius RAG Chatbot - Amazon Bedrock Professional Services Implementation

The attached diagram illustrates features of the architecture that are aligned with the AWS Well-Architected Framework.

**Generative AI**

The solution is built on an Amazon Bedrock Agent, powered by a frontier large language model (LLM) such as Anthropic Claude. The agent uses an Amazon Bedrock Knowledge Base to look up information about your organization. Safety protocols are implemented on Amazon Bedrock Guardrails and tested at scale using Amazon Bedrock Evaluations.

**Custom Indexing**

The Amazon Bedrock Knowledge Base is trained on the data sources you select, and it runs in your AWS account, keeping your data under your control. The knowledge base index is prepared and updated using native synchronization capabilities of Amazon Bedrock Knowledge Base, and optionally, custom crawlers implemented by Qalius to securely access the data you authorize.

**Enterprise Features**

Highly scalable AWS services such as AWS Web Application Firewall and Amazon API Gateway implement authorization and protection at the edge before requests are proxied to the Amazon Bedrock Agent. Amazon CloudWatch Logs record performance metrics and agent actions to support audit and continuous improvement. The overall solution accommodates the unique characteristics of your sources with the enterprise features for oversight and governance.

# Implementation

The project is implemented in five simple phases.

**Preparation**

The joint project team identifies source documents to be indexed in the knowledge base, and any custom crawlers or custom authentication mechanisms that may be required to access the source data. The team will select guardrail policies and plan automated and manual evaluation. Qalius may be provisioned least-privilege access to your AWS accounts and source control.

**Development**

Qalius provisions AWS services for the Amazon Bedrock Knowledge Base and configures the indexing of data sources. Qalius will provision an Amazon Bedrock Agent with a frontier LLM and connect it to the Amazon Bedrock Knowledge Base. Amazon Bedrock Guardrails are applied, and initial smoke tests assess the retrieval augmented generation (RAG). Custom code, if required, is prepared for the frontend interface that will be embedded in your website, intranet, or web application.

**Safety Configuration and Testing**

Engineering evaluation is followed by assessment at scale powered by Amazon Bedrock Evaluations to guide iteration of the Agent, Knowledge Base, and Guardrail configurations. Customers may supplement the automated testing with user-acceptance testing to further validate critical safety features.

**Deploy**

The project team deploys the RAG chatbot to production by activating the chat interface. Scheduled or event-driven updates of the Amazon Bedrock Knowledge Base index are activated in the background.

**Sustain**

The project team will meet for post-launch reviews to assess the performance. Token consumption and knowledge base index costs are reviewed to optimize the trade-offs between cost and performance. Optionally, you may engage Qalius for ongoing support of this RAG chatbot and your other workloads in AWS.

All services are delivered in Canada by AWS-certified specialists.