Issues with Setting Up Vector Search RAG Corpus in Google Cloud Vertex AI

Other Questions ai employee
1

I’m having trouble setting up a RAG corpus in GCP Vertex AI and connecting it to vector search. I’ve tried several different methods but keep running into problems.

First attempt - Can’t create corpus with vector search details:

{
  "name": "Document corpus",
  "info": "Corpus for document retrieval with vector search",
  "vectorConfig": {
    "vertexVectorSearch": {
      "searchIndex": "projects/my-project/locations/us-central1/indexes/doc-index-123",
      "endpointPath": "projects/my-project/locations/us-central1/indexEndpoints/endpoint-456"
    },
    "embeddingConfig": {
      "vertexEndpoint": {
        "modelPath": "projects/my-project/locations/us-central1/publishers/google/models/text-embedding-005"
      }
    }
  }
}

This should work but the API call fails completely.

Second attempt - Empty vector search config:

{
  "name": "Document corpus", 
  "info": "Corpus for document retrieval",
  "vectorConfig": {
    "vertexVectorSearch": {},
    "embeddingConfig": {
      "vertexEndpoint": {
        "modelPath": "projects/my-project/locations/us-central1/publishers/google/models/text-embedding-005"
      }
    }
  }
}

This creates the corpus but it gets stuck in INITIALIZATION status and never becomes ready.

Third attempt - Different embedding model:

I also tried using text-embedding-003 instead of 005 but that didn’t help either.

Has anyone successfully created a RAG corpus with vector search? What am I missing here?

2

had the same issue a while ago. you should ensure that both the vector search index and the endpoint are created prior to your corpus config. also verify that your service acct has the necessary permissions for vertex ai and vector search apis. when it hangs, often it’s a missing IAM role.