[mercari GEARS 2025] The Invisible Backbone: AI-Native Observability for Modern Platforms

Transcript

1. The Invisible Backbone: AI-Native Observability for Modern Platforms
 Vaibhav Khurana
 Mercari / Engineering Manager: Platform

2. Vaibhav Khurana
 
 
 Bangalore, India 
 Engineering Manager: Platform
 Spent over a decade working in areas of Platform, SRE, DevOps and Security.
 Table Tennis Badminton Travelling

3. Agenda
 
 Problem
 
 
 Approach
 
 Solution
 

4. The Problem

5. Observability as a Hurdle
 Time to Market 
 Complexity 
 Management
 Multiple checkpoints 
 
 No set format 
 
 Manage via Code
 Learning curve
 
 
 Query building 
 Regular updates
 
 
 Change of queries

6. Fragmented Visibility
 Dependent Debugging 
 Scoped Learnings 
 Limited Visibility
 
 Low Oncall Confidence 
 Learning is scoped to service.
 
 
 
 

7. Approach

8. Observability as an Enabler
 
 
 Out of the Box 
 Standard yet Personalised 
 Adaptability 
 Understands your application
 
 Configures only what’s needed
 
 Links for detailed info
 Updates with time
 
 Learn from others
 
 Central governance
 Standard setup 
 
 Holistic defaults, no base configuration 
 
 No additional code
9. Solution

10. Phase 1: Detection
 *Logos are trademarks of their respective owners.

11. Phase 2: Action
 *Logos are trademarks of their respective owners.

12. Let’s See the Results!!

13. Monitors/Alerts: A different Case!
 • Defaults works for most but not for all.
 
 • Different Thresholds
 
 • Environment specific configurations
 
 • Notification Mechanisms

14. Cherry On Top with AI

15. Why It Matters?
 
 
 Context 
 Standardisation 
 Prompt 
 Consistency and Relevance is the key
 Better Prompts for all alerts with central runbooks

16. It’s Always the team!!
 Ankush Saini
 Praveen Sai Krishna
 
 Satyam Soni
 Rajesh Sharma

17. Thank You!
 vaibhavkhurana93/