AI BriefingKakao Applies AI to Monitoring Solution 'MATRIX' to Automate Incident Analysis and Code Tuning
Key point
Ensures stability and security through alternative model chaining during LLM failures and tokenization of sensitive data.
Details
Kakao has applied AI technology to its proprietary monitoring solution MATRIX to enhance operational intelligence and flexibility. MATRIX AI serves as a bridge connecting various LLM models and is provided as a 'Role'-based service that performs specific tasks such as incident management, error log analysis, and change tracking.
Enhanced Stability and Security
To address the instability of LLM services, a chaining structure was adopted that allows configuring one or more fallback LLMs for a single role. Additionally, when calling external LLM APIs, sensitive internal system information such as SQL queries or JDBC URLs is replaced with random tokens, and reverse-mapped upon response to prevent data leakage. All requests, responses, and token usage are logged in ElasticSearch for cost prediction and quality monitoring.
Key AI Use Cases
- Incident AI Report: Analyzes APM data to go beyond simple alarms by providing root cause analysis and remediation measures. It profiles complex issues such as slow SQL or network latency to generate structured reports containing summaries, transactions, and action plans.
- Change Tracking: AI automatically monitors service status before and after deployment. It detects anomalies by comparing data with that from 24 hours or one week prior, rather than using absolute values, and supports further application analysis through structured responses in JSON format.
- AI On-call: Developed to complement environments where 24/7 monitoring for 365 days is difficult. It includes functionality to directly contact responsible personnel via phone API and TTS based on severity.
- Code Tuning Agent: Plans to pass error stack traces or Slow SQL information to coding agents via an MCP server to automate code modifications and PR generation.
MATRIX AI aims to reduce developers' repetitive tasks and improve service stability by delegating many tasks in the monitoring, detection, root cause analysis, and remediation stages to AI.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.