Building a RAG Pipeline with Qdrant and Gemini

Retrieval-augmented generation (RAG) grounds a language model in your own data, so answers stay accurate and current without retraining. Here is the pipeline I reach for.

The four stages

1. Chunk the source documents into passages.

2. Embed each chunk into a vector.

3. Store the vectors in Qdrant for fast similarity search.

4. Retrieve the top matches and pass them to Gemini as context.

5. Build the Docker Image: Use the Dockerfile to build the Docker image for the Express application.

   docker build -t express-lb .

Request Workflow Diagram

graph TD;
    A[Client] -->|HTTP Request| B[Nginx Load Balancer :8000];
    B -->|Round Robin<br/>Primary| C[App Instance 1 :4500];
    B -->|Round Robin<br/>Primary| D[App Instance 2 :4501];
    B -->|Backup<br/>if 1 & 2 down| E[App Instance 3 :4502<br/>Backup];
    C --> F[Response];
    D --> F;
    E -->|Only if needed| F;
    F --> A;

Searching with Qdrant

Qdrant returns the nearest neighbours for a query embedding in milliseconds:

const hits = await client.search('articles', {
    vector: queryEmbedding,
    limit: 5,
});

Feed those passages into the prompt and let the model answer from them. The result is grounded, cite-able output that keeps improving as your corpus grows.

Request Workflow Diagram

graph TD;
    A[Client] -->|HTTP Request| B[Nginx Load Balancer :8000];
    B -->|Round Robin<br/>Primary| C[App Instance 1 :4500];
    B -->|Round Robin<br/>Primary| D[App Instance 2 :4501];
    B -->|Backup<br/>if 1 & 2 down| E[App Instance 3 :4502<br/>Backup];
    C --> F[Response];
    D --> F;
    E -->|Only if needed| F;
    F --> A;

VS Code Laravel