Architecture diagram

Multi-agent System Process Diagram

This diagram illustrates the workflow of a multi-agent research system. Here’s how the process unfolds:

• The user submits a query to the system.
• The system creates a LeadResearcher agent.
• The LeadResearcher initiates the “Narrative Research Process,” which involves:
• Planning the research approach.
• Retrieving relevant context.
• Creating subagents for different aspects of the research (e.g., Subagent1 for aspect A, Subagent2 for aspect B).
• Each subagent may perform web searches and internal reasoning.
• Subagents complete their tasks and synthesize results.
• The system checks if more research is needed; if so, the loop continues, otherwise, it exits.
• Once research is complete, the LeadResearcher compiles the results.
• The CitationAgent processes the documents to insert citations.
• The final research report, complete with citations, is returned to the user and persisted in memory[1].

GenAI Chatbot Architecture

This architecture diagram shows how a generative AI chatbot is deployed on AWS Cloud:

• The user interacts with the chatbot via an API call secured with TLS.
• The request goes through the internet to a Load Balancer, protected by a Web Application Firewall (WAF).
• The Load Balancer routes requests to containers managed by AWS App Runner.
• The containers interact with various AWS services:
• Polly for text-to-speech.
• Bedrock for generative AI capabilities.
• Translate for language translation.
• Bhashini APIs for additional language support.
• The frontend is managed by AWS Amplify (Chatbot UI), which communicates with Kendra (search) and DynamoDB (database) for storing and retrieving data[2].

Perplexity Multi-Agent Voice Assistant Architecture

This diagram outlines a voice-enabled multi-agent assistant using AWS services:

• The user can input via microphone or chat.
• If using a mic, Amazon Transcribe converts speech to text.
• The input is sent to a Perplexity server (Docker) and processed by Bedrock MultiAgent with inline agents.
• A Supervisor Agent (powered by Amazon Nova Pro) coordinates sub-agents and filters harmful content using Bedrock Guardrails.
• Sub-agents include:
• Itinerary Agent
• Food Agent (with access to Google APIs via AWS Lambda)
• Places Agent
• ElevenLabs API is used for converting text output to speech.
• The output is sent back to the user as text or speech.
• The entire application is built using Amazon Q Developer, and frontend images are powered by Bedrock Nova Canvas[3].

Architecture & Request Flow (Travel Booking Example)

This diagram shows a travel booking application architecture using AWS:

• Frontend Application:
• User authentication via Amazon Cognito.
• UI served through AWS Amplify and CloudFront.
• Security via AWS WAF.
• Image assets managed by Nova Canvas and S3.
• Backend Services:
• API Gateway connects the frontend to backend Lambda functions.
• Multiple Bedrock agents handle specific tasks (hotel reservation, flight booking/search, etc.).
• Lambda functions manage hotel and flight bookings, searches, and airport code lookups.
• All booking data is stored in a Travel Database (likely DynamoDB).
• Amazon Cognito is also used for backend authentication.

When starting at the highest level of an enterprise’s architecture, the Conceptual level is most appropriate. This level focuses on broad, high-level views of the architecture, outlining fundamental principles, key business objectives, and the overall structure without delving into detailed activities.