Postgres as a Control Plane: Design Patterns for Real-Time Compute Offload
Presented by:
Sweta Vooda
Sweta Vooda is a Software Engineer at Sigma Computing, where she works on backend platform team and query compilers. She is also a recent Georgia tech graduate with a keep intrest in database internals. Her most recent adventure was building pgvector-remote, a Postgres extension that bridges native SQL to Pinecone's vector search engine. She is on LinkedIn at linkedin.com/in/sweta-vooda
No video of the event yet, sorry!
As Postgres continues to evolve beyond a traditional database into a control plane for orchestrating external compute systems, developers building extensions face a new class of engineering challenges. From vector search engines like Pinecone to ML inference runtimes and streaming platforms, the need to offload compute while preserving SQL simplicity and ACID guarantees is growing. This talk presents practical patterns and architectural lessons from the development of
pgvector-remote, a compute-offloading extension built at Georgia Tech. It covers key design decisions around buffering and backpressure, predicate pushdown, connection management, and transactional consistency. Attendees will gain a reusable design blueprint for building control-plane extensions that integrate with external engines while maintaining Postgres performance, simplicity, and consistency. Ideal for engineers exploring real-time analytics, search, or ML use cases directly from SQL.
- Date:
- 2026 April 23 09:00 PDT
- Duration:
- 50 min
- Room:
- Winchester
- Conference:
- Postgres Conference: 2026
- Language:
- Track:
- Postgres Extensions Day
- Difficulty:
- Medium