SaaS Multi-Agents Workshop

This workshop demonstrates how to build a multi-tenant, multi-agent architecture for a SaaS platform called "SmartResolve" - a generative AI-powered intelligent resolution engine for technical support.

Architecture Overview

The solution implements a multi-tenant architecture with the following components:

1. Knowledge Base: A pooled knowledge base with tenant isolation through metadata filtering
2. Agents: Multiple specialized agents for different tasks
3. Tenant Isolation: Implemented through metadata filtering and session attributes

Tenant Structure

The solution supports multiple tenants, with the following example tenants:

1. ClearPay (FinTech): A company that provides payment and transaction reconciliation services for small to mid-sized financial institutions.
2. MediOps (HealthTech): A platform offering EHR (Electronic Health Records) infrastructure and automation tools for medium-sized healthcare providers.

Data Structure

Each tenant has the following data structure in Amazon Simple Storage Service (Amazon S3):

tenant-id/
  ├── logs/
  │   └── microservice-logs.json
  ├── kb/
  │   └── [knowledge-base-documents].md
  ├── resolutions/
  │   └── [resolution-documents].md
  ├── sops/
  │   └── [sop-documents].md
  ├── meeting-notes.txt
  └── [tenant]_Error_Codes.txt

Technical Support Agent

The technical support agent provides the following capabilities:

1. Knowledge Base Search: Search the knowledge base for relevant information
2. Error Code Lookup: Get details about specific error codes
3. Log Analysis: Search logs for specific patterns or time periods
4. Issue Management: Create, view, and resolve technical support issues

Multi-Agent Architecture

The solution uses a multi-agent architecture with the following components:

1. Orchestrator Agent: Coordinates the work of specialized agents
2. Knowledge Base Agent: Queries the knowledge base for relevant information
3. Log Agent: Analyzes logs to identify issues
4. Code Agent: Generates code to fix issues

Tenant Isolation

Tenant isolation is implemented through the following mechanisms:

1. Amazon S3 Object Metadata: Each object in Amazon S3 has a tenant_id metadata field
2. Session Attributes: The tenant ID is passed as a session attribute to agents
3. Amazon DynamoDB Partitioning: Data in Amazon DynamoDB is partitioned by tenant ID
4. AWS Identity and Access Management (IAM) Policies: IAM policies restrict access to tenant-specific resources

Knowledge Base

The knowledge base contains the following types of documents:

1. Error Codes: Documentation of error codes and their resolutions
2. SOPs: Standard Operating Procedures for routine operational activities such as incident response, system maintenance, and data backup procedures
3. Resolution Documents: Documentation of past issue resolutions
4. Knowledge Base Documents: General knowledge base articles

Mock Data

The solution includes mock data for the following tenants:

1. ClearPay (FinTech):

   • Error codes related to payment processing
   • SOPs for transaction handling
   • Resolution documents for common payment issues

2. MediOps (HealthTech):

   • Error codes related to EHR systems
   • SOPs for patient data handling
   • Resolution documents for common healthcare IT issues

Getting Started

To deploy the solution:

1. Run the AWS Cloud Development Kit (AWS CDK) deployment script
2. Provision tenants using the tenant provisioning script
3. Generate mock data for each tenant

Workshop Labs

The workshop consists of the following labs:

1. Lab 1: Multi-tenant RAG architecture

   • Introduction to multi-tenant RAG
   • Querying the knowledge base with tenant isolation

2. Lab 2: Multi-tenant multi-agents architecture

   • Introduction to multi-agent systems
   • Orchestrating multiple agents for multi-step troubleshooting scenarios, such as analyzing logs, querying knowledge bases, and generating resolution recommendations

3. Lab 3: Tenant isolation

   • Implementing tenant isolation through metadata filtering
   • Implementing tenant data protection with enhanced security features through IAM policies

4. Lab 4: Cost per tenant

   • Tracking and analyzing costs per tenant
   • Implementing cost optimization strategies including resource right-sizing, usage monitoring, and automated scaling policies