Scaling MLOps for Computer Vision

The development of traditional apps differs from how AI products are built, primarily due to distinctions in inputs (program vs data) and process structure (sequential vs iterative). However, ML/Data Science teams could benefit from adopting established Software Engineering patterns; bridging the gap that frequently impedes the transition of ML applications to production. In this talk, David will introduce Flyte, an open-source platform that empowers ML/DS teams to collaborate effectively and expedite the delivery of production-grade AI applications.

Recogni develops high compute, low power, and low latency neural network inference processors with unmatched performance amongst automotive inference processing systems. Fabio will share how ML engineers and researchers at Recogni leverage Flyte as part of their internal developer platform to speed up machine learning experiments for Recogni’s ML-silicon co-design, to develop a state-of-the-art automotive perception stack, and to compress and mathematically convert these models for deployment.

Blackshark.ai is analyzing satellite imagery of the whole planet with AI. In this talk we explore the lessons learned from training and executing AI models at scale. It touches upon challenges such as managing large datasets, ensuring model reliability, and maintaining system performance. We will also discuss the importance of efficient workflows, robust infrastructure, and the need for continuous monitoring and optimization.