CXMind System Architecture Flow Description

CXMind is an intelligent monitoring and analysis solution built for modern Customer Experience (CX) and communication middleware. This system is designed based on a high-performance, distributed cloud-native infrastructure, balancing open-source accessibility with powerful dual-mode parallel processing capabilities.

Macroscopic Modules and Boundary Division

At the architectural level, the entire platform is divided into three core boundaries:

1. Data "Ingestion and Pre-parsing" Phase —— Ingestion Engine (IE)

   • Acts as the "mouth" and "ears" of the system.
   • The Community Edition supports standard SIP port mirroring packet capture or lightweight HEP encapsulation protocols to horizontally dock with softswitch clusters like FreeSWITCH, Asterisk, and Kamailio.
   • This layer is responsible for signaling filtering and session correlation identification.
   • (Pro Commercial Edition feature only: Supports real-time decryption and on-demand recording of massive RTP media streams).

2. Data "Storage, Arbitration, and Distribution" Hub —— Redis & Storage Layer

   • Adopts loosely coupled design ideas: State flows (Ringing, Answered, Hung up) processed by the IE no longer initiate HTTP polling to backend business applications, but are pushed to the Redis Streams / Pub/Sub event bus.
   • Persists all long-term structured call data and Call Detail Records (CDR) to disk.
   • Designed for large-scale high concurrency, ensuring "not a single drop of call records is lost" even if frontend business modules are down for maintenance.

3. Data "Response and Governance" Backend —— App Server (AS) & Admin UI (AU)

   • App Server: A powerful middleware microservice based on Node.js, responsible for organizing role-based permissions, multi-dimensional data querying and statistics, and picking up real-time events from the Redis bus, processing them, and broadcasting to online browsers across the network.
   • (Pro) Intelligent Orchestration Scheduler: Also the "brain" that triggers third-party Large Language Models (LLM) for quality inspection judgements and real-time suggestions.
   • Admin UI: A cyberpunk-styled frontend application with end-to-end encrypted responses, allowing administrators to gain a command center-level panoramic view.

User-Perceivable Data Flow

graph LR
    A[PBX (FreeSWITCH / Kamailio)] -->|SIP Signaling (UDP/TCP/TLS) | B(Ingestion Engine)
    A -->|HEP Monitoring Stream| B
    B -->|Formatted/Denoised Signaling| C[(Redis High-Speed Event Bus)]
    B -->|Long-Term CDR| D[(Persistent Storage)]
    C -->|Sub Consumption| E[App Server Side]
    E -->|WebSocket Live Broadcast| F[Browser Admin UI]

Future-Oriented Architecture Extensibility

• Zero-Intrusive Capture: Bypassing capture from switches without changing the existing call center or PBX topology network, adding no pressure to business nodes.
• Horizontal Seamless Scaling: The Ingestion Engine can seamlessly scale out horizontally in K8s, cooperating with cloud Load Balancers to share massive traffic peaks (a single node's commercial limit breaks thousands of concurrent routes).
• Complete Privatized Isolation: Every line of code regarding data cleansing and intelligent judgement runs within a local VPC, providing the highest level of assurance for compliance and privacy policies (PCI-DSS and other standards).