AWS re:Invent 2025 - Ripple: Building an intelligent, multi-agent system for 24/7 operations-IND3301

Ripple: Building an Intelligent, Multi-Agent System for 24/7 Operations

Overview

• Ripple, a fintech company, has built an AI-powered operations platform to transform their platform operations using AWS services.
• The goal is to leverage AI and automation to improve the monitoring and troubleshooting of the XRPL (XRP Ledger) blockchain network.

Challenges

• Reliance on C++ experts to make sense of the large volumes of debug logs from the decentralized XRPL network.
• Difficulty in quickly identifying and resolving issues due to the complexity of the peer-to-peer network.
• Need for a more scalable and efficient way to process and analyze the massive amounts of operational data.

Solution Architecture

The solution consists of several key components:

Log Processing Pipeline

• Raw logs from XRPL nodes (validators, hubs, client handlers) are ingested into Amazon S3 using GitHub workflows and AWS Systems Manager.
• An S3 event trigger invokes a Lambda function to parse the log data, extract relevant chunks, and store them in Amazon CloudWatch.

Code Analysis Pipeline

• The Rippled and Standards repositories are automatically synced using Amazon EventBridge.
• The Git repository processor pulls the latest changes, versions the code and documentation, and stores them in Amazon S3.
• A knowledge base injection job then ingests this data into an Amazon Neptune graph database.

Graph-Based Code Analysis

• The code and documentation data stored in Neptune is used to build a lexical graph, capturing relationships between entities like functions, classes, and modules.
• This graph-based approach enables efficient retrieval and analysis of the codebase, leveraging the Bedrock knowledge base and re-ranking capabilities.

Multi-Agent Platform

• The core of the solution is a multi-agent platform built using the Strand SDK, an open-source framework for coordinating AI agents.
• The platform consists of four AI agents:
  1. Orchestrator Agent: Receives user queries, classifies the intent, and coordinates the execution of specialist agents.
  2. Code Analysis Agent: Leverages the graph-based knowledge base to provide insights about the XRPL codebase.
  3. Log Analysis Agent: Performs operational analytics on the CloudWatch log data, using a CloudWatch query generator agent to construct accurate queries.
  4. CloudWatch Query Generator Agent: Generates optimized CloudWatch queries based on predefined patterns and instructions.

Prompt Engineering and Model Integration

• The AI agents use carefully crafted system prompts to define their roles, responsibilities, and behavioral guardrails.
• The agents are powered by large language models, with the flexibility to use different models for different tasks (e.g., a lighter model for the Orchestrator, a more capable model for the specialist agents).
• The agents use the Model Context Protocol (MCP) to interact with external systems like Amazon CloudWatch.

Demonstration and Key Capabilities

• The solution is demonstrated through a chat-based interface, where users can ask questions about the XRPL network and receive detailed, correlated responses.
• Examples include:
  • Identifying the number of proposals received by a UNL (Unique Node List) validator from other peers.
  • Correlating log messages to understand the sequence of events during a consensus round.
• The solution provides enhanced data troubleshooting and processing capabilities, reducing the time and effort required to investigate operational issues.

Lessons Learned and Future Opportunities

• Importance of context engineering and prompt design for effective AI-powered solutions.
• Benefit of removing the bi-directional dependency between platform engineers and C++ experts.
• Potential to expand the solution's capabilities, such as:
  • Leveraging agent memory and identity management features in Amazon Bedrock.
  • Implementing transaction-level monitoring and analysis to detect and mitigate network-level issues like spam and scams.

Technical Details

• AWS services used: Amazon S3, Amazon CloudWatch, Amazon Neptune, Amazon EventBridge, Amazon API Gateway, Amazon Cognito, AWS Lambda, AWS Secrets Manager, AWS Systems Manager Parameter Store.
• AI frameworks and tools: Strand SDK, Bedrock Agent Core, Claude language model.
• Key metrics: 900+ XRPL nodes, 100 million ledgers processed, 0.4 cents per transaction fee.

Business Impact

• Improved operational efficiency and reduced time to investigate and resolve issues in the XRPL network.
• Empowered platform engineers to quickly analyze logs and codebase without relying on C++ experts.
• Potential to expand the solution's capabilities to address additional operational and security challenges in the XRPL ecosystem.

Conclusion

Ripple's AI-powered operations platform demonstrates the power of leveraging AI and automation to tackle complex, decentralized infrastructure challenges. By building a multi-agent system on AWS, Ripple has been able to improve the monitoring and troubleshooting of their XRPL network, leading to enhanced operational efficiency and reduced reliance on specialized expertise.