Managing nodes with Vector Property

Neo4j Graph Platform
1

We are using KNN like similarity search capabilities using one of the Vector databases. Now we are looking at integrating that capability in to our graph database on Neo4j. I see that KNN based vector search is supported in Neo4j. But it looks like we can not generate the vectors externally and store into Neo4j. We would like the following process flow. In each node we will have only one property which is used for this functionality..

  1. While loading / updating the data into neo4j for selected nodes we want to store the embeddings of one of the property (generated using our own model outside of neo4j).
  2. Load / update the vector property in to neo4j
  3. As part of our regular query make use of the neo4 provided similarity search.
    Looking at the documents looks like we need to integrate any LLM model for embeddings into neo4j, this will not work for for us.
    Any suggestions / recommendations to follow to achieve the same. Especially generating the embeddings outside and storing into neo4j, then follow the standard procedure / process?.
2

You can generate vectors externally by using embedding calls to any of the 4 embedding providers:

  • OpenAI
  • Azure OpenAI
  • Vertex AI
  • AWS Bedrock
3

Thank you for your response, I have see this document. This expects that we configure our GenAI engine into neo4j.

What I am looking for is as follows:
With the integration, I need to call the following cypher to generate the embeddings:

MATCH (n:Tweet { id: 1234 })
RETURN genai.vector.encode(n.text, "VertexAI", { token: $token, projectId: $project }) AS embedding

Instead is there a way I have my embedding generated like:
in my client code I will generate the vector and get the vector property.

gen_vector = [1234,876,890,321,987] ->vector generated using my own model.
Then:

"MERGE (n:Tweet {id:1234,tweet: 'Test Tweet', vector: $gen_vector}) "

Is this type of data population is possible and then may be I can create index on vector property and use in the queries?