PostGIS Day 2025 Talk Now Live

Last week, I had the honor of presenting at PostGIS Day 2025 on how we built geospatial embeddings directly into Geobase.app. The talk is now available to watch on YouTube, and I'm excited to share it with you.

About the Talk

In this talk, I share how we built geospatial embeddings directly into Geobase.app, our geospatial backend. By extending PostGIS with PGVector, we enabled semantic search and similarity queries over maps, rasters, and vector datasets.

The talk focuses on:

• How to use PostGIS with PGVector — Practical integration approaches and techniques
• Benchmarks — Performance metrics and comparisons for geospatial similarity search
• Practical considerations — Real-world implementation details, challenges, and solutions
• Similarity search use cases — How to leverage geospatial embeddings for similarity queries

We walk through the technical design choices from embedding generation to storage and retrieval, with a focus on similarity search capabilities. The session highlights lessons learned in integrating modern AI/ML methods with a Postgres-native stack, making advanced retrieval workflows accessible to developers and analysts.

Key Takeaways

• PostGIS + PGVector integration — Practical approaches to combining spatial capabilities with vector similarity search
• Benchmarks and performance — Performance metrics and optimization strategies for geospatial similarity queries
• Storage and retrieval — Efficient ways to store and query geospatial embeddings in Postgres
• Similarity search — How to implement and use similarity search with geospatial data
• Practical considerations — Real-world implementation details, challenges, and solutions
• Lessons learned — Insights from building production-ready geospatial AI systems

Introducing GeoEmbeddings Service

Building on the work presented in this talk, we're excited to announce the launch of GeoEmbeddings — the first in our suite of GeoAI features.

GeoEmbeddings is a specialized service that enables you to store, search, and query embeddings alongside their geographic footprint, capture time, and rich metadata—all in one place. Unlike traditional vector databases that don't understand space and time, GeoEmbeddings bridges the gap between geospatial data and AI embeddings.

What makes GeoEmbeddings special

• 📍 Spatially-aware storage — Link embeddings to point, line, or polygon footprints
• 🕑 Time of capture — Perfect for satellite and drone data with temporal context
• 🔎 Hybrid queries — Filter by area, time, or properties and then re-rank results with vector similarity
• 📦 Bulk ingest — Upload millions of embeddings via Parquet, CSV, or NDJSON
• ⚡ High-performance indexes — Powered by PostGIS + pgvector, tuned for scale
• 🇪🇺 EU-based hosting — Data stored and processed entirely within the European Union, meeting GDPR and data residency requirements

Join the Private Beta

We're launching GeoEmbeddings as a private beta to ensure we can provide the best possible experience for our early users. During this phase, we'll be gathering feedback, providing dedicated support, and iterating quickly based on your needs.

If you're interested in learning more about GeoEmbeddings or want to join the private beta, check out our announcement post for details on how to get started.

Watch the Full Talk

You can watch the complete talk on YouTube:

Thank you to the PostGIS Day 2025 organizers for the opportunity to share our work, and to everyone who attended the session. We're excited to continue pushing the boundaries of what's possible with geospatial AI and PostGIS.

If you have questions about the talk or want to discuss geospatial embeddings, feel free to reach out to us at hello@geobase.app or join our Discord community.