Hari Chandrasekhar (VP of Engineering Core @ Sedai) spent years building Sedai's core data pipeline, the system that pulls billions of data points a minute from thousands of cloud resources.
At that scale, failure is inevitable, so Hari redesigned the pipeline to handle it.
On this episode of 1 IDEA, Hari breaks down:
- The hub and spoke redesign that centralized retry decisions
- How Sedai runs production jobs on spot instances
- Why smaller components make AI code generation more accurate
CHAPTERS:
00:00 Introduction
01:22 Why failure is inevitable at scale
06:20 What broke in the old architecture
11:22 Why smaller codebases make AI more accurate
12:05 The hub and spoke redesign
21:57 Running production jobs on spot instances
28:34 How to validate two systems at once
38:02 Where Sedai uses AI beyond coding
39:41 Prevent every failure vs. tolerate it