Practical AI Machine learning in your database
5 snips
Jun 22, 2022 
Montana Low and Lev Kokotov, co-founders of PostgresML, share their journey in enhancing machine learning capabilities within the Postgres database. They discuss real-world applications from Instacart, emphasizing the importance of practical, simpler predictive models over complex systems. The duo highlights the shift to Postgres, its advantages in integrating machine learning, and the innovative solutions they developed for performance and scalability. They also unpack the user-friendly features of PostgresML, aiming to make machine learning more accessible for data scientists.
AI Snips
Chapters
Transcript
Episode notes
Instacart's Elasticsearch Struggle
- Montana Low's team at Instacart initially used Elasticsearch to scale their product catalog.
- This approach eventually led to performance issues due to growing document sizes and real-time data requirements.
Read-Time Joins
- Joining data at read time, rather than index time, significantly reduced workload and improved performance.
- This realization led to the exploration of Postgres as a potential solution.
Pandemic-Induced Switch to Postgres
- During the pandemic, Instacart's Elasticsearch cluster overloaded, leading to timeouts.
- This crisis forced them to switch to their Postgres prototype, revealing further optimization needs.