MLOps.community

Ethan Rosenthal from Runway discusses challenges of multimodal AI, transitioning from language to video models, and tools for content creators. He explores managing large datasets, transitioning research to production, and collaboration with cloud infrastructure tools. Accelerating research and utilizing resources efficiently at startups are also covered.

Boris Selitser, Co-Founder of Okareo, discusses the power of synthetic data and agent evaluations in LLM development. Topics include safeguarding AI products, online evaluation in agent systems, custom evaluations, and the role of synthetic data in AI development. The conversation also explores agent architectures, challenges faced by startups, and the potential of agent frameworks in the tech industry.

Discussion on challenges in deploying AI models in manufacturing, optimizing models for corner cases, transitioning models to production, and exploring trust in traditional ML vs LLMs. Special guests share insights on managing battery health, semiconductor manufacturing, and maintaining reliability and customer trust in industrial settings.

Miguel Fierro, Principal Data Science Manager at Microsoft, discusses the challenges of applying ML in robotics and the integration of computer vision in sports analytics. He highlights the role of AI in strategic game analysis and explores the evolution of recommendation systems, emphasizing the importance of real-time architectures for personalized recommendations.

Demetrios Brinkmann, AI strategist at Uber, discusses the evolution of Michelangelo platform at Uber, from basic ML predictions to deep learning and generative AI. Covering challenges faced in early versions and improvements in Michelangelo 2.0 and 3.0 like Pytorch support, enhanced model training, and integration of technologies like Nvidia’s Triton and Kubernetes. The platform now includes features like a Genai gateway, compliance guardrails, and model performance monitoring to streamline AI operations.

Join Kamran Khan and Matthew McClean as they discuss AWS Trainium and Inferentia, powerful AI accelerators offering enhanced performance and cost savings. They delve into integration with PyTorch, JAX, and Hugging Face, along with support from industry leaders like W&B. Explore the evolution and performance comparison of these AI chips, flexibility in model training with Trainium, and workflow integration with SageMaker. Discover the distinctions between inference and training on accelerators and explore AWS services for generative AI.

Benjamin Wilms, a chaos and resilience engineering expert, discusses integrating Chaos Engineering into the CI/CD pipeline for system resilience. They explore the cultural shift needed to embrace failures as learning opportunities and the transition to structured chaos engineering experiments. The conversation also covers reflection on errors, AWS's Trainium, chaos engineering methods, and the intersection of chaos engineering and observability for reliable systems.

Tom Smoker, Cofounder of WhyHow.ai, discusses using knowledge graphs in multi-agent systems. Topics include mitigating hallucination issues, optimizing search with knowledge graphs, agile problem-solving, and integrating vector databases. The conversation explores agents in multi-agent systems, stepping back for growth, and using AI models for automated content creation and revenue generation.

Dave Nunez, Partner at Abstract Group, discusses how AI is changing developer documentation strategies. He emphasizes the need to rewrite the developer education playbook for AI-focused software. Topics include enhancing user experience with design cues, escape patches, and intuitive documentation. The importance of clear, user-friendly content and effective onboarding experiences is highlighted.

Cody Peterson, Senior Technical Product Manager at Voltron Data, discusses the importance of open standards in MLOps. Topics include challenges with scalability in data tools like Pandas, leveraging the Ibis project for big data processing, and the power of Apache Arrow in data systems. The conversation also covers transitioning between platforms, considerations for data system selection, and future plans for the Ibis project.