Since Uber embarked on its AI journey back in 2015, AI has become an integral part of every facet of Uber’s business. Michelangelo is an end-to-end AI/ML platform that democratizes machine learning and enables ML practitioners to seamlessly build, deploy, and operate AI/ML solutions at Uber’s scale. In this talk, we will share the past, present and future of our Michelangelo platform and how it is evolving to support both predictive and generative models.

In the world of organizations striving to be more data-driven, the standard software stack isn’t enough. We need to be able to access relevant data, process it “in time” and keep track of what we have done at the speed of a page load. When this information is coming from multiple clouds, and your organization is migrating and ever-improving its data-infrastructure, architecting, building and delivering solutions that scale is not always straight forward. In this talk, I discuss some of the considerations when working in a multi-cloud environment with near-real time latencies.

Prima, a data-driven insurance leader, encountered significant challenges in sustaining growth and maintaining consistent feature engineering practices across its expanding customer base. With over 2.5 million customers and a footprint in multiple European regions, the pressure to innovate and streamline operations became paramount. This talk delves into Prima’s journey from using an in-house centralized feature solution, which posed considerable operational challenges, to adopting Tecton’s Enterprise Feature Platform. Attendees will gain insights into the selection process, the advantages of a git-centric feature platform, and the tangible results of this transformation – including improved efficiency, reduced latency, and enhanced collaboration between data scientists and engineers. Through this case study, learn how embracing the right tools in MLOps can significantly alter the trajectory of business outcomes.

This talk explains how we integrated Tecton into our company’s CI/CD tools. We built a system that allows developers to use familiar tools within our company to quickly setup features for their products and maintain the high security standards our customers depend on.

At this talk we will describe our journey taking HelloFresh from proven data science projects to mature machine learning products. We will discuss how we developed our MLOps strategy, how we selected our tool stack, from feature store to model monitoring, and how we use our system to train 1500+ machine learning models per week. We will conclude with our takeaways on what we would do differently if we had to start again.

This fireside chat with Databricks CEO and Co-Founder Ali Ghodsi will focus on LLMs, real-time, and Other trends in the production ML space.

Join us for a talk on the journey of scaling ads ranking at Pinterest using innovative machine learning algorithms and innovations in the ML platform. This presentation will showcase the transition from traditional logistic regressions to deep learning-based transformer models, incorporating sequential signals, multi-task learning, transfer learning. Throughout the process, we encountered various challenges and gained valuable lessons. Discover the hurdles we overcame and the insights we gained in this talk, as we share the transformation of ads ranking at Pinterest and the lessons learned along the way.

In this panel, ML practitioners gather together to share their approaches toward building recommender systems. They will also discuss the biggest technical challenges their organizations faced when deploying recsys at scale, common pitfalls, and mistakes to avoid when implementing production-ready recsys.

In this live Tecton demo, Matt will show how you can build a real-time AI application to detect fraud with nothing more than Python, Using Tecton’s new compute engine, we will define and test features entirely locally before productionizing them for real time inference using GitOps best practices