Matt Cornillon


I’m a database, OpenSource and Postgres lover for as long as I can remember. I am currently working as a Solution Architect for Aiven, a company deeply involved in the OSS community. Also, I’m helping the organization of the local Postgres Lille Meetup and this year I’ll be part of the PGDay France organization (which will be held in Lille in June!)

Talk Description: How I found my Pokémon cards thanks to Postgres: an AI journey

When it comes to data storage and retrieval, PostgreSQL offers more than just traditional tabular data management. This session explores the exciting potential of combining PostgreSQL with AI technologies, illustrated through a relatable example—finding and identifying specific Pokémon cards. While Pokémon cards serve as a fun and engaging example, the techniques presented have wide applications in various data-driven domains such as object detection and autonomous driving.

We'll begin with an overview of the machine learning pipeline for image recognition where we’ll discuss how a Convolutional Neural Network (CNN) can be trained to recognize Pokémon cards, generating embeddings that capture essential visual features.

Our main focus will be on pgvector, an extension for PostgreSQL dedicated for managing vector data like embeddings efficiently. We will cover how to store such embeddings, set up indexes for quick retrieval, and conduct similarity searches right within your PostgreSQL database, and at scale!

If you're intrigued by the integration of AI with PostgreSQL for real-time, high-performance data operations, this session will offer both theory and hands-on practices to get you started.