Abstract

We explore critical considerations for developing and deploying coding agents at scale in a production environment that has generated over 250k commits to date. We delve into architectural decisions, including interactivity, multi-agent systems, orchestration, state management, security, effective tool design, etc. We will also take a deep dive into our Evaluation Driven Development approach we follow at Jules.

About the Speaker

Dr. Mehadi Hassen is a Staff Research Engineer at Google with a noteworthy career launching high-impact, LLM-powered products and over a decade of experience working at the intersection of machine learning, coding, and computer security. He currently serves as the Modeling and Research Technical Lead for Jules (a remote coding agent), where he directs the research roadmap and technical execution. Throughout his tenure at Google, Mehadi has played a pivotal role in driving major LLM product launches. He co-designed and introduced code editing as a core capability within Gemini models and spearheaded the development of the code generation and editing platform that powers Google Colab, Android Studio, and Chrome Dev Tools.
Earlier in his career, he was a founding member of Google’s internal AutoML platform, contributing to core ML infrastructure used by teams across Gmail, YouTube, and other products. Mehadi holds a Ph.D. in Computer Science from the Florida Institute of Technology, where he studied under Dr. Philip Chan, and is a graduate of the Stanford LEAD program at the Stanford Graduate School of Business. His current research interests center on LLM post-training, coding agents, and the security of LLM-powered products.