Data science for real problems
summary
A data science team builds personalization as narrow microservices deployed to Google Cloud Platform using containers, Kubernetes, Helm and GitHub CI, training with Airflow and a managed Hadoop job and storing models/telemetry in BigQuery and Bigtable; services expose REST endpoints, run 24x7 on SRE on‑call, and collect RPS, latency and click‑through metrics to enable A/B and multi‑armed bandit testing with automated traffic routing and ~20‑minute streaming freshness.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:15Topics
- 0:44About LaLa Digital
- 1:26Personalization recommendation project
- 5:27Basic workflow
- 6:34Continuous deployment
- 7:02Monitoring and tracing
- 7:54Datadriven team
- 8:41Experiments
- 9:21Case study
- 10:26Data Flow
- 11:12Lessons Learned
- 15:05Data sources
related talks
37mBuilding a Data Platform from scratch
29mData Science in Production
44mBig Data Science on Google Cloud
34mData Science at PMI - The Tools of The Trade
23mBuilding Scalable, Observable MLOps Systems on Google Cloud
41mModular & Reusable: A Recipe to Scaling Data Across a Portfolio
37mBig Data Analytics, Machine Learning, Real Time Processing
35mBuilding a data analytics platform on strong opinions held loosely