#60 Large Scale Production Engineering India (LSPE-IN) meet

Event Agenda and Speaker info

10:45 - Registration

11:00 - Tea

11:15 - Keynote - LSPE for Collaboration : Jessica Prabhakar, Site Lead, Yahoo India & Debansu Saha, Organizer #LSPE-IN

11:30 - Yahoo Mail on Cloud : Bhavin Shah, Sr. Manager and Senthil Ponnuswamy, Principal Engineer, Yahoo India.

Operating a globally recognized email platform requires serving hundreds of millions of users while maintaining strict sub-second SLAs across petabytes of stateful data. Migrating this massive ecosystem from legacy, on-premise infrastructure to a modern cloud-native paradigm is a complex engineering feat. This talk explores the architectural evolution of Yahoo Mail as it transitions to the cloud. It will cover zero-downtime data migration strategies, multi-region fault tolerance, and the containerized design choices powering high-concurrency workloads. Attendees will gain practical insights into modernizing stateful core systems to support resilient operations and next-generation, AI-driven capabilities at scale.

Bhavin is a Senior Manager in Yahoo Mail Engineering with 20 years of experience in large-scale distributed systems. He leads the Core Mail Data engineering group, focusing on cloud migration, scalability, and reliability for Yahoo Mail’s critical backend infrastructure.

12:15 - FinOps and Cloud Efficiency in days of token : Arkadip Basu, Principal Architect, Platform Engineering

As AI moves into production, tokens have emerged as a fundamental unit of compute economics. This talk explores how FinOps must evolve to manage model selection, inference costs, context windows, and agent workflows. We will examine practical strategies to measure cost per request, task, and business outcome, alongside real-world patterns like prompt optimization, model routing, and caching. Learn how treating tokens as a first-class engineering metric enables teams to balance response quality, latency, and reliability without blowing up infrastructure budgets.

1:20 - Lunch

2:20 - Sahastrabaahu: Scaling Local LLMs to the Thousands Requests :
Mitesh Sing Jat, Principal Architect, Sarvam

Scaling local LLMs like Ollama for privacy and cost control often founders on GPU memory limits, high concurrency demands, and single points of failure. This talk introduces Sahastrabaahu—a high-performance Golang hierarchical model router designed to move beyond single-server deployments. We will explore the architecture required to orchestrate 100+ local Ollama nodes, detailing how Sahastrabaahu handles 10,000+ requests per minute with dynamic load distribution, fault tolerance, and low-latency routing across distributed enterprise infrastructure.

3:00 - The 800 MB Lie - What Your Docker Image Says About Your Architecture : Dr. Sachin Garg, Founder

Docker images often become unnecessarily bloated when applications depend on full OS distributions and dynamic libraries. Static binaries and minimal images like "FROM scratch" can dramatically reduce size, dependencies, and attack surface. A case study demonstrates a mail-sync container under 30 MB. The broader lesson: minimize unnecessary infrastructure, dependencies, and microservices to build simpler, safer, more maintainable systems.

Dr. Sachin Garg has 27+ years from embedded systems to AI/ML platforms. He is a regular LSPE contributor and has also presented at Open Source North America, Open Source India, and various other events.

3:30 Open forum, Tea, snacks and networking.

Parking directions

Enter into the EGL park from Koramangala Indiranagar intermediate ring road. The second building on your left is the venue, Yahoo!. This is the Torrey Pines building. Ask the security for help and tell them that you have come for the #LSPE-IN event at Yahoo and they will help you park and entry.