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.
35:35
forum Ask episode
web_stories AI Snips
view_agenda Chapters
auto_awesome Transcript
info_circle Episode notes
question_answer ANECDOTE
Uber's Ad-Hoc ML Development
Before Michelangelo, Uber's ML development was ad-hoc and fragmented, using various open-source tools.
This limited ML's impact due to short timeframes and a lack of standardized pipelines.
insights INSIGHT
Lack of Project Tiering
Michelangelo 1.0 lacked project tiering, treating all projects equally regardless of business impact.
This meant less important projects received the same resources as high-ROI ones.
volunteer_activism ADVICE
Optimizing ML Use Cases
Uber encouraged optimization of ML use cases after seeing significant ROI.
This led to adopting more advanced techniques like deep learning and a plug-and-play architecture.
Get the Snipd Podcast app to discover more snips from this episode
Uber's Michelangelo: Strategic AI Overhaul and Impact // MLOps podcast #239 with Demetrios Brinkmann.
Huge thank you to Weights & Biases for sponsoring this episode. WandB Free Courses - http://wandb.me/courses_mlops
// Abstract
Uber's Michelangelo platform has evolved significantly through three major phases, enhancing its capabilities from basic ML predictions to sophisticated uses in deep learning and generative AI. Initially, Michelangelo 1.0 faced several challenges such as a lack of deep learning support and inadequate project tiering. To address these issues, Michelangelo 2.0 and subsequently 3.0 introduced improvements like support for Pytorch, enhanced model training, and integration of new technologies like Nvidia’s Triton and Kubernetes. The platform now includes advanced features such as a Genai gateway, robust compliance guardrails, and a system for monitoring model performance to streamline and secure AI operations at Uber.
// Bio
At the moment Demetrios is immersing himself in Machine Learning by interviewing experts from around the world in the weekly MLOps.community meetups. Demetrios constantly learns and engages in new activities to get uncomfortable and learn from his mistakes. He tries to bring creativity into every aspect of his life, whether analyzing the best paths forward, overcoming obstacles, or building Lego houses with his daughter.
// MLOps Swag/Merch
https://mlops-community.myshopify.com/
// Related Links
From Predictive to Generative – How Michelangelo Accelerates Uber’s AI Journey blog post: https://www.uber.com/en-JP/blog/from-predictive-to-generative-ai/
Machine Learning Education at Uber // Melissa Barr & Michael Mui // MLOps Podcast #156: https://youtu.be/N6EbBUFVfO8
--------------- ✌️Connect With Us ✌️ -------------
Join our slack community: https://go.mlops.community/slack
Follow us on Twitter: @mlopscommunity
Sign up for the next meetup: https://go.mlops.community/register
Catch all episodes, blogs, newsletters, and more: https://mlops.community/
Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/
Timestamps:
[00:00] Uber's Michelangelo platform evolution analyzed in podcast
[03:51 - 4:50] Weights & Biases Ad
[05:57] Uber creates Michelangelo to streamline machine learning
[07:44] Michelangelo platform's tech and flexible system
[11:49] Uber Michelangelo platform adapted for deep learning
[16:48] Uber invests in ML training for employees
[19:08] Explanation of blog content, ML quality metrics
[22:38] Michelangelo 2.0 prioritizes serving latency and Kubernetes
[26:30] GenAI gateway manages model routing and costs
[31:35] ML platform evolution, legacy systems, and maintenance
[33:22] Team debates maintaining outdated tools or moving on
[34:41] Please like, share, leave feedback, and subscribe to our MLOps channels!
[34:57] Wrap up
MLOps.community 