Presented by:

E722def5fcbe21573a0a831a86ea64b3

Sachin Khanna

from AWS

I am Database Architect/DBA with extreme passion for postgreSQL. I am into Database Performance tuning, Troubleshooting and Optimisation. I love PostgreSQL and its flavours including AWS Aurora PostgreSQL. I am having a wide range of Development, Implementation and administration experience with big organisations using database like Oracle, DB2 LUW and PostreSQL. As a senior member of My Team, I’m able to combine my love for PostgreSQL Databases and Other open source database like yougabyte, casandra db , mongodb etc. I am always looking forward for new additions in PostgreSQL and always open to contribute to PostgreSQL community.

Checkout the slides

With the rise of generative AI, vector embeddings have become the key to enabling powerful applications like large language models. However, storing and searching vector data is not feasible option for traditional databases.

In this presentation, we will explore pgvector - a PostgreSQL extension that enables efficient storage and retrieval of vector data. We will cover:

  • What are vector embeddings and why are they critical for generative AI models like LLMs? We'll look at how vector data is generated and structured.

  • Managing vectors in PostgreSQL with pgvector - we'll demo with sample vector data and using pgvector's vector operators like distance and similarity search.

  • Optimizing search with IVF and HNSW indexes - we'll compare two indexing approaches available in pgvector for fast approximate nearest neighbour search.

  • Benchmarking vector query performance - we'll showcase pgvector's query speed on indexed vector data and discuss scaling considerations.

  • Building a demo app with pgvector - we'll walk through a simple generative AI demo app powered by vectors stored and queried from PostgreSQL.

By the end of this session, you'll understand the role of vector data in modern AI and how PostgreSQL can be augmented with pgvector to manage vectors for productive generative AI applications. The demos will highlight pgvector's capabilities and how it enables PostgreSQL to support vector-based workloads.

Date:
2024 February 29 - 12:20
Duration:
40 min
Room:
Grand Victoria A
Conference:
PGConf India, 2024
Language:
Track:
Application Developer
Difficulty:
Medium

Happening at the same time:

  1. YDB — Adding PostgreSQL compatibility to a Distributed SQL database
  2. Start Time:
    2024 February 29 12:20

    Room:
    Robusta + Arabica

  3. Postgres load balancing is secretly broken: The cancellation problem
  4. Start Time:
    2024 February 29 12:20

    Room:
    Grand Victoria B