Transforming Cable Network Reliability with Agentic AI & Graphs

The Journey to Autonomous Networks

Telco Autonomous Network Journey

• Telco autonomous network journey structured by TM Forum with different phases:
  • Manual operation: Managing thresholds, rules, and hard-coded scripts
  • Closed-loop based operation: Network can sense data, understand it, reason on it, and take actions

Requirements for Autonomous Networks

1. Data: Extracting and preparing network data (KPIs, telemetry, configurations, etc.)
2. Knowledge Fabric: Transforming raw data into usable formats (time series, dependencies, embeddings, etc.)
3. Predictive Intelligence: Applying analytics, machine learning, deep learning, and foundation models
4. Autonomous Execution: Deploying agents to take actions based on insights

AWS Services and Frameworks

• Execution and Agentic AI:
  • Transagent: Open-source framework for creating and managing agents
  • Agent Core: AWS's comprehensive agentic platform for deploying agents at scale
• Automated Reasoning:
  • Bedrock: AWS's framework for automated reasoning using large language models
• Purpose-Built Databases:
  • S3 Vector: Cost-effective solution for unstructured network documentation
  • Amazon Neptune: Graph database for storing and querying network topology
  • Graph Analytics: Performing graph algorithms for network analysis
  • Graph Deep Learning: Applying deep learning techniques on network graph data

Autonomous Network Use Cases

1. Orchestration, Governance, and Planning:
   • Agents triggered by network events to understand and respond to issues
   • Guardrails and evaluation to ensure secure and authorized agent access
2. Observability and Proactive Manner:
   • On-demand observability to troubleshoot, recommend, and assess network health
   • Proactive event analysis, prioritization, and recommendations
3. Root Cause Analysis and Service Impact:
   • Leveraging network dependencies for change management and service impact assessment
4. New Use Cases:
   • Automating data engineering tasks like topology discovery and feature engineering
   • Accelerating the adoption of AI in network operations

Cox's Journey to Autonomous Networks

Cox Communications Overview

• Multiple service operator providing data, video, and voice services to over 6 million customers
• 180,000 miles of hybrid fiber coax and fiber-to-the-home networks
• 20,000 employees, with 100 in the Network Analytics and Reliability Enablement team

Transitioning from Reactive to Proactive Network Operations

• Historically relied on customer calls to diagnose and troubleshoot issues
• Shifted focus to harnessing data to understand systems, processes, and policies
• Aimed to reduce customer calls and truck rolls by addressing root causes

Service Health Ecosystem

1. Network Health: Aggregating and correlating SNMP probe data with network topology
2. Node Health: Combining geospatial information, time-series telemetry, and online/offline traps
   • Classifying events as urgent, critical, or impaired
   • Prioritizing labor based on customer healthy minutes
3. Premise Health: Differentiating between outside plant and in-home issues

Results and Impact

• 22% reduction in call volumes
• 10% reduction in truck rolls
• 48% reduction in impaired customer minutes

Digital Twin and the Service Health Platform

Constructing the Digital Twin

1. Discovering Network Assets: Automating the collection and integration of asset data from multiple sources
2. Data Quality Checks: Ensuring the accuracy and completeness of the network topology
3. Federated Telemetry: Aggregating alarms, KPIs, and customer transactions into a real-time representation of the network

Leveraging the Digital Twin

1. Analytics and Insights: Applying machine learning and graph algorithms to the digital twin data
2. Agentic AI Integration: Connecting the digital twin to agent-based systems for automated reasoning and actions

Strands Agent Architecture

• Routing Lambda manages agent requests and responses
• Agents scale out to process tasks from SQS queues
• Open Search used as a vector store for caching
• Leveraging Strands and Bedrock as the core components

Agent Core Enhancements

1. Agent Core Gateway: Provides a standardized communication platform for agents to access various tools and capabilities
2. Agent Core Memory:
   • Short-term memory for current incident context
   • Long-term memory for institutional knowledge and continuous improvement
3. Agent Core Observability:
   • Visibility into agent performance and cost
   • Tracing agent workflows and correlating with business outcomes

Key Takeaways

1. Data is Everything: There are no shortcuts - building high-quality, high-velocity data sets is crucial.
2. Intentional Innovation: Fostering a startup mentality with cross-functional collaboration.
3. Adopt and Go: Being willing to forge new ground and adapt as needed.
4. Building for Change: Designing systems and processes with adaptability in mind.

Technical Details and Business Impact

• Cox Communications operates a 180,000-mile hybrid fiber coax and fiber-to-the-home network
• Employs 20,000 people, with 100 in the Network Analytics and Reliability Enablement team
• Achieved a 22% reduction in call volumes, 10% reduction in truck rolls, and 48% reduction in impaired customer minutes
• Leveraging AWS services and frameworks, including:
  • Amazon Neptune for graph database
  • Amazon Open Search for event storage and analytics
  • Strands and Bedrock for agent-based systems
• Developed a digital twin of the network, combining topology and telemetry data
• Integrated agentic AI capabilities, including:
  • Automated root cause analysis and recommendations
  • Proactive event detection and prioritization
  • Observability and cost optimization for agent-based systems

The key business impact is the transformation from a reactive, customer-driven network operations model to a proactive, data-driven approach that prioritizes reliability and customer experience. By harnessing the power of data, analytics, and agentic AI, Cox has been able to significantly improve operational efficiency and customer satisfaction.