2026-09-22 –, Main Stage
I've built banking pipelines for 10 years. My company just told me we're deploying AI agents. I started researching what changes — and found 5 things my pipeline has no answer for:
- "Build passed" — but the model wasn't tested against this week's data
- "Health check passed" — but "up" doesn't mean "correct" for AI
- "Deployment successful" — but which version: the code, the model, or the context?
- "Cost normal" — but token costs don't appear in your cloud bill like compute does
- "No alerts" — but the AI failed silently because nobody defined what "wrong" looks like
5 failure modes that your current pipeline has no answer for. 5 slides each. 5 minutes. Drawn from production failures documented in public post-mortems and infrastructure gap analysis from 10 years of banking deployments.
A fast, opinionated 5-minute format: one failure mode per minute, drawn from documented production post-mortems and infrastructure gap analysis. The structure maps cleanly to the Ignite format — 4 slides per failure mode (the assumption, why it breaks for AI, what it costs, and the right question to ask instead). No tools, no demos — just the 5 mental model shifts every DevOps engineer needs before their first AI deployment. Designed to spark open space conversations rather than close them.
Saideep Navakoti is a Senior DevOps Engineer with 10 years of experience building and operating cloud-native banking infrastructure on Kubernetes. His career has been entirely in financial services — designing CI/CD pipelines, deployment automation, and observability systems for platforms handling critical financial transactions. SaiDeep holds AWS Solutions Architect and CKA certifications, and is co-author of published research on context engineering for AI agents (arXiv:2603.14057). Currently focused on the intersection of platform engineering and AI/ML workloads in regulated banking environments. Based in Atlanta, GA.