A Spring Boot application built with GraalVM that demonstrates semantic search using Oracle Vector Store and OpenAI embeddings.

GitHub: alina-yur/oracle-database-vector-search

Find full details in a blog post.

• GraalVM, or install with sdk install java 25.0.1-graal
• Oracle Database with Vector support (for example, Autonomous Database 26ai)
• OpenAI API key

By default, this project uses Oracle Autonomous Database via TLS connection.

Required environment variables:

export OPENAI_API_KEY=
export DB_PASSWORD=
export ORACLE_JDBC_URL='jdbc:oracle:thin:@(description=(address=(protocol=tcps)(port=1522)(host=<your-adb-host>))(connect_data=(service_name=<your-service-name>))(security=(ssl_server_dn_match=yes)))'

Optional environment variable:

export DB_USERNAME=ADMIN

If you start with only the (description=...) descriptor body, prepend jdbc:oracle:thin:@.

mvn spring-boot:run

The application automatically loads sample pet store inventory data on startup.

Now let's build it with Native Image:

mvn -Pnative native:compile
➜ eza -l ./target | grep "store"
... 107M ... vector-pet-store
./target/vector-pet-store

You can search for the pet store items using natural language:

curl "http://localhost:8080/petstore/search?query=Treats%20for%20small%20loud%20dogs"
# or with httpie
http localhost:8080/petstore/search query=="find cat food with tuna"

Example requests:

• "Find dog toys" -> finds "Labrador Bark Control Chews", "Heavy Duty Rope for Large Breeds"
• "Find cat food" -> finds "Gourmet Tuna Souffle for Cats", "Gourmet Chicken Soup for Cats"

The API returns semantically similar products using vector similarity search.

Using Docker or Podman:

podman run -d   -p 1521:1521   --name oracle-free   -e ORACLE_PASSWORD=mypassword   -e APP_USER=appuser   -e APP_USER_PASSWORD=mypassword   -v oracle-data:/opt/oracle/oradata   gvenzl/oracle-free:latest

Point ORACLE_JDBC_URL at the local database and run the app.

• Startup: ~1.5 seconds
• Package size: ~107 MB standalone executable
• Memory: Significantly lower

Native Image compiles your Spring Boot application ahead-of-time into a self-contained executable with faster startup (typically 20-30x), lower memory footprint (typically 3-5x), and compact deployment.