How we built a scalable end-to-end architecture for real-life machine learning
summary
Describes a cloud-based end-to-end architecture that runs many live ML models handling audio, speech, text and numeric data, supporting multiple training and inference lifecycles and both supervised/unsupervised and traditional/neural approaches. Covers how security and privacy were incorporated into the cloud design, how the system scales training and inference, and how software is packaged and operated in a polyglot Java/Python environment.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:11Gong: Business Context
- 0:41The Gong Product
- 1:12Understanding Spoken language
- 2:51Sample Business Insight
- 3:18Built-in Models and User Assisted Models
- 5:09Everyone's Process
- 6:10Challenges
- 11:08Research Environment: a separate AWS Account
- 12:40Security First: Bastion Hast
- 13:28The Export Process 1. Use Bastion Host
- 14:11Denormalization makes it easier (new)
- 15:09Research vs. Engineering Responsibilities
- 15:45Two Very Different Types of Models
- 16:47Packaging Python into Dockers
- 19:52Running a Machine Learning Model
- 22:49Operations Research
related talks
11mEngineering Scalable ML Infrastructure for Modern Apps
40mBuilding highload ML powered service
19mBuilding Resilient ML Cloud Integrations
04mInterview
36mMulti-Cloud Machine Learning Data and Workflow with Kubernetes
05mAI and ML are changing traditional business
41mInfrastructure for ML Applications
30mEssential Ingredients in Scaling Organizations for ML