100K Embedded Properties. Just Add Queries.

Test SQL Server 2025 vector search without the embedding step.

You may have read in earlier blog posts that I put 100,000 synthetic property listings with pre-generated embeddings on GitHub. You can test SQL Server 2025 vector search without paying for your own embedding calls.

The database is called SemanticShoresDB. Each property has a 1,536 dimension embedding generated from its description using OpenAI’s text-embedding-3-small. I also pre-embedded 900+ search phrases so you can immediately run queries.

I built a Python wrapper app with Claude Code to demo it. You search for something like “waterfront” or “pet friendly home” and it converts that to a vector. Then it compares against all 100,000 property embeddings using cosine distance.

The query uses SQL Server’s VECTOR_DISTANCE function. This is brute force search hitting all rows. The execution plan shows exactly what you’d expect when scanning 100,000 rows.

You don’t need the wrapper app I demonstrate here to learn about this. Restore the database and write queries directly. Check execution plans. Experiment with different distance functions. The embeddings are already there.

Video walkthrough: