Several People Are Tuning: Data and Machine Learning at

on YouTube ↗29m

" summary

Slack has two years of public data, a newly formed machine learning team (six months) and a seven‑person data engineering team that builds a modern stack on S3/Parquet with a Hive metastore, Kafka + Secor, Scoop and Spark/Presto and moves ETL into Airflow. The first deploys are a simple channel recommendations system using item‑to‑item similarity, linear regression and k‑NN, served as a daily static Hulk model key‑value for real‑time use. Experimentation (A/B) and hiring experienced ML engineers drive product metrics rather than toy models.

AI-written summary · about

" chapters (12)

1. 0:00Introduction
2. 3:52Invest in data infrastructure
3. 7:00Data team challenges
4. 9:07Early data warehouse strategy
5. 13:18Hiring in data culture
6. 16:49Friction from experienced hires
7. 18:53OODA loop and ML work
8. 21:19ML for product problems
9. 23:24Minimum viable data products
10. 25:49Channel recommendations
11. 27:11Batch models and inspirations
12. 28:55Conclusion

" related talks

• 11mRecent trends in data and machine learning technologiesBen Lorica · Strata Data Conference 2019 - New York, NY
• 30mOperationalizing Machine LearningPallav Agrawal · TWIMLcon: AI Platforms 2019
• 32mTeam Teardown: Evolving an ML PlatformSam Charrington, Jing Huang, Shubhi Jain, Jolene Zhou · TWIMLcon: AI Platforms 2019
• 19mSustaining ML in the enterpriseBen Lorica · Strata Data Conference 2019 - London, United Kingdom
• 35mBuilding Data-Intensive TeamsElias Nema · Berlin Buzzwords 2021
• 22mReal-world applications of machine learning: Beyond the hypeRon Miller, Fred Sadaghinani, Ilia Papas, William Fong · Collision 2017
• 24mJosh Wills, Head of Data Engineering, SlackJosh Wills · MLconf SF 2017
• 42mTales from the Field: The Future is NowEllen Friedman · Jfokus 2019