14–15 Nov 2026
Bengaluru
Asia/Kolkata timezone

Scaling Local AI: Benchmarking vLLM and Building Personal Agents

15 Nov 2026, 14:50
1h
Workshop Room

Workshop Room

Workshop (60 mins) Artificial Intelligence & Machine Learning (AI/ML)

Speaker

Sayed Imran
66degrees

Description

As AI moves from massive data centers to local and edge environments, developers need practical, open-source ways to deploy efficient, high-throughput models. This workshop provides a hands-on guide to building personal AI agents tailored for workplace automation, served via vLLM on Ubuntu and powered by modern unified memory architectures.

Before writing any agentic code, we will break down the mechanics of unified memory by explaining how CPU and GPU memory pooling drastically reduces data transfer bottlenecks and accelerates inference on accessible hardware like the NVIDIA DGX Spark.

From there, we will transition into a live, practical workshop focused on scale and efficiency. We will deploy vLLM, exploring how its implementation of PagedAttention and continuous batching solves memory fragmentation and allows for high-throughput serving at scale. To validate this, we will perform live benchmarking on the deployment to measure real-world token generation speed, latency, and memory utilization.

Finally, we will consume this optimized local inference server to build and wire up personal AI agents capable of automating routine development and operational tasks. Attendees will experience the excitement of actively querying their own agents against the DGX Spark, seeing firsthand how serving requests at the local level transforms and accelerates daily workflows. You will leave with a clear understanding of how to benchmark and serve LLMs efficiently, along with the code necessary to spin up your own local AI agents.

Summary

This hands-on workshop explores how to efficiently serve local LLMs using vLLM on Ubuntu. Attendees will learn the mechanics of unified memory, perform live benchmarking to measure high-throughput inference, and build personal AI agents for workplace automation

Speaker Bio

Golden Kubestronaut, 7x Google Cloud Certified, Cloud Native BBSR Organizer, Speaker & Mentor

Speaker LinkedIn / Socials https://www.linkedin.com/in/sayed-imran/
Difficulty Intermediate

Presentation materials