AWS Architecture Blog·April 23, 2026Modernizing KYC with Event-Driven Serverless Architecture and Agentic AI on AWS
This article details a modern, cloud-native architecture for Know Your Customer (KYC) processes, moving away from monolithic, batch-processing legacy systems. It leverages an event-driven serverless approach on AWS, integrating agentic AI to automate and accelerate real-time identity verification, fraud detection, and compliance validation. The design emphasizes scalability, low latency, and dynamic adaptation using a supervisor-sub-agent model.
Read original on AWS Architecture BlogFinancial institutions face increasing pressure to modernize Know Your Customer (KYC) processes due to rising transaction volumes, regulatory complexity, and demand for instant customer onboarding. Traditional monolithic KYC systems struggle with latency, availability, and scalability, relying on batch processing and manual handoffs. This leads to higher operational costs and hinders real-time compliance.
Cloud-Native Event-Driven Architecture for KYC
The proposed architecture transforms KYC into an autonomous, intelligent, and real-time process using event-driven principles and serverless AWS services. It's designed to process high-volume KYC validations securely within minutes, handling sensitive PII data while maintaining strict security and regulatory compliance across geographies.
- Amazon MSK (Managed Streaming for Apache Kafka) forms the communication backbone, enabling asynchronous, real-time message exchange between AI components and enterprise systems.
- AWS Lambda provides serverless compute for event listeners and integration layers, consuming events from MSK, invoking AgentCore asynchronously, and publishing results.
- Amazon Bedrock hosts foundation models for AI tasks like OCR, language processing, and behavioral analysis.
- Amazon Bedrock AgentCore provides native orchestration, session management, and memory persistence for AI agents.
Agentic AI Orchestration Layer
At the core is the KYC Orchestration Supervisor Agent, built on Amazon Bedrock AgentCore. This intelligent coordinator dynamically determines optimal sub-agent collaboration patterns, analyzing case characteristics (document types, geography, risk indicators) to construct context-aware execution plans. It delegates tasks to specialized sub-agents:
- Identity Verification Agent
- Document Analysis Agent
- Fraud Detection Agent
- Compliance & Risk Agent
- Customer Experience Agent
Key Architectural Advantage
Unlike traditional rule-based systems, this agentic approach provides dynamic adaptation and autonomous decision-making. AgentCore's built-in session state management, shared memory, and automatic context preservation across asynchronous workflows are crucial for maintaining coherence in complex, multi-agent interactions. Confidence scores from sub-agents guide the supervisor, allowing for automatic approvals (high confidence), additional verification (medium), or human review (low).
Agents utilize KYC Knowledge Bases, powered by Amazon OpenSearch Serverless and Amazon S3, for grounded information from internal policies and regulations. Integration with existing on-premises systems is handled via AgentCore Gateway using OpenAPI schemas and Lambda-based integrations, ensuring secure and authorized access through AgentCore Identity.