Dushyant Behl

Short CV: Dushyant is a software engineer in IBM Research. As part of his current job, he has contributed to and led the development of many products ranging from eBPF based networking and observability solutions, blockchain interoperability protocols and applications and IBM AIOps based closed loop assurances for 5G Network Functions. He has diverse set of background in developing appliction backends and scalable products. His current interest lie in developing AI Platforms for IBM WatsonX.ai suite of products with a special focus on fine tuning large language models.

Title of Talk 1: ebpf for Networking: Warp Speed Kernel Networking

Synopsis: In this talk, we delve into the transformative power of eBPF (extended Berkeley Packet Filter) in revolutionizing kernel networking. eBPF has emerged as a game-changer, offering unprecedented flexibility and efficiency in network programming within the kernel space. We will explore how eBPF enables developers to deploy custom networking functionalities with remarkable performance gains, propelling network applications to warp speed. From traffic filtering to complex packet manipulation, eBPF empowers network engineers to achieve previously unattainable levels of scalability and agility. Join us as we navigate through the fundamentals and advanced applications of eBPF for networking, uncovering the keys to unlocking warp speed kernel networking.

Title of Talk 2: ebpf for Observability: Observing the kernel with a thousand paper cuts.

Synopsis: In this presentation, we embark on a journey into the realm of observability through the lens of eBPF (extended Berkeley Packet Filter). With its granular tracing capabilities, eBPF empowers us to dissect the kernel's behavior with a myriad of fine-grained insights - akin to observing a complex organism through a thousand paper cuts. We will explore how eBPF facilitates real-time monitoring, profiling, and troubleshooting of kernel-level activities, unveiling hidden intricacies and uncovering elusive performance bottlenecks. From pinpointing latency spikes to unraveling intricate system interactions, eBPF equips us with the tools to achieve unparalleled visibility into the inner workings of the kernel. Join us as we navigate the landscape of observability with eBPF, leveraging its power to decode the kernel's mysteries one paper cut at a time.

Title of Talk 3: Fine Tuning LLMs: What, How and Why

Synopsis: This session provides a concise exploration of fine-tuning large language models, focusing on what it entails, how it's done, and why it's crucial. We cover the motivations behind fine-tuning, key methodologies, and considerations for optimizing model performance. This session aims to offers valuable insights into leveraging fine-tuning to maximize the effectiveness of large language models across various applications.