Vivek of Roku, Inc.

In the world of AI and machine learning, most discussions stop at model development — but that’s only half the story. Based on real-world experience, I’ve seen a persistent gap between building ML models and deploying them into production, where they can deliver actual value. This talk, “From Notebooks to Production: ML in Practice,” aims to bridge that gap.

We’ll start with a practical case study—classifying spam emails—to walk through the end-to-end ML lifecycle, including data preparation, training, evaluation, deployment, and monitoring. Along the way, we’ll explore the real challenges teams face when transitioning from experimentation to production, such as concept drift, scalability, and cross-functional collaboration.

The session dives into MLOps practices and tools that enable production-grade AI systems, including CI/CD pipelines, model versioning, monitoring, and agile team workflows. You’ll learn how to structure collaborative ML workflows, operationalize models reliably, and scale AI systems without falling into common traps.

This session is ideal for ML engineers, data scientists, software developers, and in-fact anyone interested in practical, field-tested approaches to building and deploying machine learning systems at scale.

Biography:

Vivek Bharti is a Senior Machine Learning Engineer at Roku, where he works across the entire machine learning lifecycle — from developing advanced Deep Learning and ML models in the Natural Language Understanding (NLU) domain to deploying scalable solutions that serve millions of users.

With 8 years of experience, Vivek has led end-to-end ML projects involving predictive modeling, NLP, computer vision (CV), and broader AI applications. His expertise spans model development, model deployment, monitoring, and continuous improvement. He is currently working on multiple patent submissions related to full-stack ML systems, covering data pipelines, model development, deployment, and maintenance.

Vivek recently delivered a guest lecture at NYU on the machine learning lifecycle and is an invited speaker at several IEEE chapters. He is passionate about mentoring and sharing practical ML insights with the broader AI/ML community

Email:

Babul of SCAN Health Plan

Autonomous agent AI systems, which are agentic AI systems that have the ability to operate autonomously, are transforming healthcare in streamlining clinical and administrative operations. Yet, calling such systems to real healthcare settings comes with problems concerning data heterogeneity, safety, regulation, and trust. As this paper analyses these barriers, it also suggests explanation-based design principles, modularity-based design principles, and human-based oversight-based design principles. we explore the approach of deployment (via hybrid RAG models, real-time orchestration and compliance layer). We use case study analysis to assess system performance: in triage, radiology, ADE detection and in dementia-care. Our results provide practical implications and a bespoke platform through which agentic AI may be securely and saleable implemented into significant healthcare processes.

Biography:

Babul Sahu is a Lead AI Engineer with over 20 years of experience in software development, including 5 years in machine learning and more than 2 years in Generative and Agentic AI. He leads all AI initiatives at SCAN Health Plan, where he built the organization’s first AI team and established its enterprise AI strategy. Babul has designed and deployed secure, production-grade AI systems—ranging from LLM-powered automation platforms to multi-agent frameworks using the framework LangGraph and CrewAI. His work includes hybrid RAG pipelines, healthcare-specific NLP, and MLOps on Azure. He is passionate about responsible AI, AI infrastructure at scale, and applying AI to solve real-world healthcare challenges.

Email: