with-agents

Architecting with Google Cloud — selected guide

How to build an AI-powered mobile ad platform

Google Cloud Tech 8 of 10
Open original playlist (opens in a new tab)
In this collection Browse 10 summaries 8 of 10

Moloco vice president of engineering Chang describes the company's 2022 mobile-ad system as a latency-bounded inference path connected to offline training, feedback, and analysis pipelines.

Key Points Covered

  • Chang explains how auction deadlines constrain feature retrieval, model complexity, and inference. [00:02:13]-[00:06:28]
  • He describes GKE bid processing, Bigtable and BigQuery features, TensorFlow Serving, and engagement feedback. [00:08:33]-[00:12:49]
  • Bigtable, Cloud Storage, BigQuery, Dataflow, and Looker serve different replay, training, and analysis needs. [00:13:54]-[00:17:01]
  • Chang attributes scaling and cost control to native autoscaling, varied node types, garbage collection, and unit-cost tracking. [00:17:01]-[00:21:17]
  • This 2022 first-party account is not a reusable reference design; validate current documentation and workloads, plus privacy, consent, profiling, fairness, safety, security, governance, and advertising-regulatory requirements.

Full video: https://www.youtube.com/watch?v=Hc5xAK0cWgA(opens in a new tab)