These AI & Data Engineering Sessions Are a Must-Attend at ODSC East 2025

As AI and data engineering continue to evolve at an unprecedented pace, the challenge isn’t just building advanced models — it’s integrating them efficiently, securely, and at scale. From LLMOps and AI agents to real-time data pipelines and enterprise AI deployment, this year’s ODSC East 2025 sessions offer practical insights from industry leaders who have tackled these challenges head-on.

Whether you’re navigating AI decision support, technical debt in data engineering, or the future of autonomous agents, these sessions provide actionable strategies, real-world case studies, and cutting-edge frameworks to help you stay ahead. Senior leaders, engineers, and AI practitioners alike will gain practical takeaways to implement in their own organizations — without getting lost in unnecessary complexity.

Real Time Delivery of Impressions at Scale

Tulika Bhatt, Senior Software Engineer at Netflix

Netflix processes 18 billion daily impressions, which fuel video ranking algorithms and real-time adaptive recommendations. To meet the growing demand for low-latency, in-session responsiveness, Netflix employs a hybrid architecture combining batch processing and real-time computations. Exponential Moving Averages (EMAs) represent impressions across time windows from 2 hours to 365 days, balancing accuracy, scalability, and cost efficiency.

This talk will explore how Netflix efficiently computes, stores, and serves EMAs through a Key-Value store, batch jobs, and real-time offset calculations. While this approach ensures sub-second response times and near real-time updates, it introduces complexities in managing dual systems, formula intricacies, and large-scale transport jobs. Attendees will gain insights into how Netflix overcomes these challenges to deliver a seamless user experience at scale.

Structuring the Unstructured: Advanced Document Parsing for AI Workflows

Cedric Clyburn, Senior Developer Advocate at Red Hat

William Caban, Product Manager at Red Hat

Modern organizations generate vast amounts of unstructured data — PDFs, scanned documents, and proprietary files — making AI-driven processing a challenge. Extracting text isn’t enough; engineers must preserve structure, context, and relationships to build effective retrieval-augmented generation (RAG) pipelines and fine-tune models.

This session explores open-source tools and techniques for transforming unstructured documents into structured formats like JSON and Markdown. Attendees will learn how to tackle challenges such as multi-page tables, image-heavy layouts, and scanned documents using context-aware methods. Walk away with practical strategies to bridge the gap between unstructured data and AI applications, improving model performance and decision-making.

Building AI’s Foundation: The Critical Role of Data Architecture

Bill Inmon, CEO of Datavox.ai

AI, ChatGPT, and ML rely on believable data, yet much of today’s data is unstable, unstructured, and unreliable. Without a solid data architecture, these technologies cannot fulfill their promise. Traditional data warehousing was built for structured data, but today’s landscape also includes textual and analog data, each with distinct challenges that structured data practices cannot address.

This session explores how modern data architectures must evolve to support structured, textual, and analog data, ensuring AI-driven technologies operate on reliable and meaningful information. Attendees will gain insights into best practices for handling diverse data types, laying the foundation for more effective AI and machine learning applications.

10 Most Neglected Data Engineering Tasks

Veronika Durgin, VP of Data at Saks

Organizations juggle high-priority projects, urgent system failures, and a hidden third category — neglected data engineering tasks that quietly accumulate into technical debt. Ignoring these “forgotten bucket” tasks leads to unplanned disruptions, inefficiencies, and escalating costs.

Join Veronika Durgin as she uncovers the most overlooked data engineering pitfalls and why deferring them can be a costly mistake. Learn how to define “done” in data projects, the importance of Data SLAs, and the hidden trade-offs of build vs. buy decisions. Walk away with strategies to proactively manage technical debt before it derails your data operations.

AI Software Engineering Agents: What Works and What Doesn’t

Robert Brennan, CEO of All Hands AI

AI in software development has had mixed results — while autocomplete tools like Copilot are widely adopted, autonomous agents like Devin and OpenHands spark both excitement and skepticism. Some engineers claim 10x productivity gains, while others see only noise and tech debt.

The key? Setting realistic expectations and understanding how to use these tools effectively. This session explores what today’s AI agents can handle, what they might excel at by 2025, and which tasks will always require human expertise. Walk away with practical strategies for integrating AI-powered development agents without compromising code quality or maintainability.

Agentic AI in Action: Build Autonomous, Multi-Agent Systems Hands-On in Python

Dr. Jon Krohn, Host of the SuperDataScience Podcast

Edward Donner, Co-founder and CTO of Nebula.io

Autonomous AI agents are transforming how we plan, collaborate, and solve complex problems with minimal human intervention. This hands-on half-day workshop will provide a deep dive into agentic AI, equipping you with both theory and practical coding experience to build and deploy multi-agent systems.

You’ll start by exploring leading agent frameworks — LangChain, CrewAI, Microsoft Autogen, OpenAI’s Swarm, and Hugging Face’s Smolagents — before designing agentic architectures based on industry best practices. The core of the session features a live coding lab, where you’ll build autonomous AI agents from scratch in Python, progressing from basic research agents to fully autonomous systems capable of decision-making.

We’ll wrap up with insights into deployment strategies, future trends, and ethical considerations, including whether AI agents will outnumber humans in the workforce. By the end, you’ll have a solid conceptual foundation and hands-on experience, enabling you to confidently implement autonomous AI in your own projects.

Beyond the Prompt: Architecting Reliable Enterprise LLM Agents

Vivek Muppalla, Director of AI Engineering at Cohere

Building scalable, safe, and seamless LLM agents for enterprise remains a challenge, despite advances in tool-calling capabilities. Enterprise applications — such as customer support agents — require strict accuracy, safety, and orchestration across multiple frameworks, tools, and models.

This talk will guide you through the key decision-making process in developing enterprise-ready AI agents. Learn how to strategically select frameworks, define robust tools, implement human-in-the-loop safety measures, and establish effective evaluation criteria. We’ll also explore how synthetic training data can enhance model performance. Walk away with actionable insights to build reliable, enterprise-grade LLM agents that meet real-world demands.

Build with AG2: Open-Source AgentOS

Chi Wang, PhD, Founder of AutoGen (Now AG2), Senior Staff Research Scientist at Google DeepMind

AI agents are transforming industries by seamlessly integrating multiple models (OpenAI, Anthropic, Gemini, open-weight providers) with diverse toolsets. This session will showcase practical, cutting-edge AI agent applications, demonstrating how to build and deploy multi-agent systems for research, customer support, and software development.

Attendees will gain a solid understanding of agent-oriented programming, learn how to construct multi-agent solutions for complex tasks, and accelerate production-readiness by 10x. Real-world use cases include financial analysis, customer support automation, game design agents, and AI-powered research assistants. Walk away with actionable insights to contribute to the AI agent ecosystem and drive innovation in your industry.

Effective AI Decision Support: Overcoming Both Human and AI Fallability

Finale Doshi Velez, PhD, Professor at Harvard University

AI is increasingly used to support human decision-making, from clinical treatment recommendations to business intelligence. The goal is for human+AI teams to outperform either alone, yet this complementarity often fails — humans may over-rely on incorrect AI outputs or dismiss accurate AI guidance.

This talk explores research-backed principles for designing AI systems that provide effective decision support. Learn how explanations impact trust and usability, how task context influences AI reliance, and how individual differences shape interaction with AI recommendations. While rooted in healthcare applications, these insights apply broadly to any field leveraging AI for decision support.

Building the Backbone: Scalable LLMOps & MLOps to Enable AI Integration Across Teams

Pablo Vega-Behar, Director of Machine Learning Engineering at Fitch Group

As AI becomes a strategic imperative, enterprises need scalable, efficient AI integration. At The Fitch Group, the Emerging Technology team has developed a reusable AI backbone, enabling multiple teams to integrate AI without major increases in headcount. This structured LLMOps/MLOps approach accelerates deployment while ensuring governance and adaptability.

This session provides senior leaders with a business-focused framework for implementing enterprise AI operations in financial services. Learn how to prioritize AI integration requests, streamline governance, and deploy reusable AI components that empower non-specialist teams. Walk away with strategies to scale AI efficiently and maximize business impact without deep technical expertise in every department.

Future-Proof Your AI & Data Strategy

AI is no longer a futuristic concept — it’s an operational necessity. However, building AI that is scalable, reliable, and strategically impactful requires more than just great models. The ODSC AI engineering sessions at ODSC East 2025 this May 13th to 15th will arm you with the tools, frameworks, and strategies to navigate AI integration, optimize data pipelines, and manage the complexities of AI engineering.

Don’t miss the opportunity to learn from top AI & data experts who have successfully built and deployed AI solutions at scale. Whether you’re an executive looking to drive AI adoption or a data engineer optimizing real-time architectures, these sessions will equip you with the knowledge and practical insights to make AI work for your organization — efficiently and effectively.