Solution Overview

Communication service providers (CSPs) manage massive volumes of data generated by billions of connected devices to ensure seamless and uninterrupted operations. To accomplish this, they rely on data-intensive network management systems that monitor critical performance metrics like latency, maintain reliability during network congestion, and uphold stringent security measures to protect against cyberattacks.

Recognizing the complexity and scale of these operations, our solution highlights the transformative power of integrating artificial intelligence and MongoDB to automate network management, significantly streamlining workflows for operators. By merging MongoDB with large language models (LLM) and adopting a retrieval-augmented generation (RAG) strategy, we have developed an AI-assisted chatbot that transforms network operations into a conversational, automated, and informed system. This chatbot taps into network logs, maintenance records, customer data, and sensor data stored in MongoDB to provide actionable insights for anomaly detection, root-cause analysis, mitigation recommendations, and more.

This approach empowers network operators to manage complex workflows with minimal human involvement, accelerating the rollout of new services and potentially increasing CSPs’ annual revenue by up to 5% through autonomous networks.

Reference Architecture

There are three main components of this solution:

1. Source data ingestion

Before querying data, network managers need to ingest log entries and telemetry events in real time, capturing details like IP addresses, geographic data, request paths, timestamps, router logs, and sensor data. This architecture leverages MongoDB's stream processing capabilities to automatically capture and process incoming data into MongoDB Atlas, creating a comprehensive view of network activity. This enriched dataset is then leveraged within our RAG architecture for decision-making.

2. Question-based data selection

The second component of this solution addresses the questions from network managers, such as “What might be causing the client’s video streaming issue in Toronto?” This query undergoes an initial processing where the LLM generates a customized aggregation pipeline to select the appropriate data for analysis. Concurrently, vector-embedded data can be efficiently retrieved through semantic search, allowing for the extraction of closely related information.

3. Inference and natural language output

Once MongoDB identifies the relevant data, a subsequent LLM translates this information into natural language explanations for the user. During this process, the LLM analyzes the retrieved data to detect patterns and anomalies, enabling precise identification of root cause candidates and supporting informed decision-making. For example, it might uncover that an overloaded local CDN node, along with high requests from older routers, are causing the problem.

Data Model Approach

This solution leverages server network logs, organized as time series data, using the following MongoDB schema:

Code Snippet

Building the Solution

This AI-powered network management solution employs a RAG framework with MongoDB Atlas to enhance data-driven diagnostics in complex telecommunication environments.

Solution guidelines

1. Clone application locally

   • Download the github repository following this link: https://github.com/mongodb-industry-solutions/telco-ai-ops/tree/main.

2. Environment configuration

   • Configure environment variables for:
     • API key from the LLM service provider.
     • MongoDB connection URI.
     • Database and collection names.

3. Application deployment

   • Install python appropriate version.
   • Create and activate a virtual environment.
   • Install dependencies.
   • Run the application.

Key Learnings

• MongoDB and gen AI transform network management: Integrating LLMs with MongoDB's capabilities like aggregation pipelines and vector search effectively streamlines network management by reducing human intervention, optimizing processes, and ultimately automating critical operations.

• Data management is critical: Network management systems produce vast amounts of data from network logs and user requests, creating substantial challenges in data interoperability, privacy, and efficient processing. Effective solutions require flexible, scalable databases that can store and handle high-frequency, complex data streams.

• MongoDB offers the tools to drive AI-powered solutions: A robust database infrastructure, combined with a flexible document model and advanced vector search capabilities, allow CSPs to efficiently develop AI applications. Specifically, vector search streamlines the retrieval of semantically relevant content, boosting the performance of LLMs that fuel chatbot technologies.

Technologies and products used

MongoDB developer data platform:

• MongoDB Atlas
• MongoDB Aggregation Pipeline
• Atlas Vector Search
• Atlas Stream Processing

Authors

• Benjamin Lorenz, Principal Industry Solutions
• Diego Canales, MongoDB