Section 7: Lab: Automatic Ticket Classification and Routing
- Design the categories, priorities and resolver groups the AI must choose from
- Compare few-shot LLM classification with an embedding-based classifier
- Measure with a confusion matrix and send low-confidence tickets to a person
- Lab: classify the sample ticket set and measure accuracy before going live
Section 8: Lab: Duplicate Tickets and the Knowledge Base
- Find similar tickets with vector search to merge duplicates of the same problem
- Suggest fixes from past tickets and knowledge base articles, citing the source
- Draft new KB articles from closed tickets for staff to review before publishing
- Lab: a service desk assistant that proposes KB answers for staff to confirm
Section 9: Lab: From Runbook to Runbook Agent
- Rewrite runbooks as structured steps with conditions, commands and expected results
- Build an agent in n8n that calls tools through MCP, starting with read-only permissions
- Let the agent pick a runbook by incident type and report each step
- Lab: an agent that diagnoses a slow database by runbook and writes the findings to the ticket
Section 10: Lab: Safe Auto-Remediation
- Choose tasks suited to automatic fixes, with command allowlists, dry runs and a rollback plan
- Ask for human approval in chat before changing a system, linked to the change process
- Keep an audit log of every decision and every command
- Lab: fix a memory leak with a restart that needs approval first
Section 11: Measuring and Governing AIOps
- Metrics: MTTA, MTTR, alerts per incident and classification accuracy
- Track LLM costs and match the model to the task
- Data policy and PDPA when tickets and logs contain personal data
- Workshop: an AIOps results dashboard in Grafana
Section 12: Workshop: Capstone from Alert to Closed Ticket
- Receive a complex simulated fault from the instructor
- Detect it, correlate alerts into an incident and open a ticket automatically
- Find the cause with the AI assistant and fix it with the runbook agent and approval step
- Close the ticket with a draft KB article and a post-incident report