Scaling AI Features: Capacity Planning and PTU Cost Realities in Azure -Dhanabal

Scaling AI Features: Capacity Planning and PTU Cost Realities in Azure
A successful AI pilot proves a feature works, but not whether it can handle production traffic or justify its cost. When we planned to scale an LLM-powered feature to a much larger audience, we had to look at tokens, throttling, and pay-as-you-go vs. provisioned throughput (PTUs). In this talk, I'll cover how we projected capacity from token usage, validated it with a small load test (including hitting HTTP 429s), and why output tokens turned out to be the biggest lever. I'll also share how the cost numbers reshaped product decisions like rollout strategy and fallback behavior.

Speaker: Dhanabal Chandrasekaran
Dhanabal has spent nearly two decades building software in the .NET ecosystem and currently works as a Principal Software Engineer at H&R Block. His interests include solving problems around distributed systems, cloud architecture, and increasingly, practical applications of AI in software development. Outside of building systems, he enjoys mentoring developers and sharing lessons learned from building software at enterprise scale.

SeeSharper: Have Dotfiles, Will Travel: Portable Developer Environments
Duane Newman (@duanenewman)
Ever find yourself setting up the same tools, shell aliases, and configurations every time you start working on a new machine or development environment? Dotfiles provide a simple way to keep your development environment consistent, portable, and version-controlled. In this lightning talk, we'll look at what dotfiles are, how to manage them with Git, and how to use them with VS Code Dev Containers to automatically configure your environment. Whether you're working across multiple machines, using WSL, or spinning up new dev containers, a little dotfile automation can save time and make your development experience more consistent.

Sponsored & Hosted By: Alien Arc Technologies
Location Provided by: Quest Analytics