Skip to content
Nappai Documentation

NVIDIA Rerank

The NVIDIA Rerank component lets you improve the relevance of search results in your Nappai dashboard. By connecting it to a retriever that pulls documents from your data store, you can send those documents to NVIDIA’s reranking API, which reorders them so the most useful ones appear first.

How it Works

  1. Connect a Retriever – First, link a retriever component that knows how to fetch documents from your data source (for example, a vector store or a database).
  2. Set the API Key – Enter your NVIDIA API key. This key authenticates your requests to NVIDIA’s cloud service.
  3. Choose the Base URL – The default URL (https://integrate.api.nvidia.com/v1) points to NVIDIA’s integration endpoint. You can change it if you use a different environment.
  4. Select a Model – Pick the reranking model from the dropdown. The current option is nv-rerank-qa-mistral-4b:1.
  5. Enter a Search Query – Type the query you want to use to search your documents.
  6. Rerank – When you run the component, it sends the retrieved documents and the query to NVIDIA’s API. The API returns the same documents reordered by relevance.
  7. Outputs – The component gives you two outputs:
    • A new retriever that already includes the reranking step.
    • The list of search results, now sorted by relevance.

Inputs

  • Retriever: Connect a retriever component that fetches documents from your data source.
  • API Key: Enter the secret key that authorizes your requests to NVIDIA’s API.
  • Base URL: The base URL of the NVIDIA API. Defaults to https://integrate.api.nvidia.com/v1.
  • Model: Choose the reranking model to use.
  • Search Query: Type the text query you want to search for.

Outputs

  • Retriever: A retriever that automatically reranks documents using NVIDIA’s API. You can feed this output into other components that expect a retriever.
  • Search Results: A list of documents returned by the reranker, ordered from most to least relevant. These can be displayed in a table, sent to a downstream processor, or used to trigger actions.

Usage Example

  1. Add the NVIDIA Rerank component to your workflow.
  2. Connect a retriever (e.g., a vector store retriever) to the Retriever input.
  3. Paste your NVIDIA API key into the API Key field.
  4. Leave the Base URL as the default unless you have a custom endpoint.
  5. Select the model (nv-rerank-qa-mistral-4b:1).
  6. Enter a search query such as “customer support tickets for billing issues”.
  7. Run the workflow. The component will return the reranked documents in the Search Results output, which you can then display in a dashboard table or pass to another component for further processing.
  • ContextualCompressionRetriever – A retriever that compresses context before searching.
  • VectorStore – Stores embeddings of documents for fast similarity search.
  • Retriever – Generic interface for components that fetch documents.

Tips and Best Practices

  • Keep your API Key secure; store it in Nappai’s secret manager instead of hard‑coding it.
  • Use concise, specific queries to get the best reranking results.
  • If you have many documents, consider limiting the number of results before reranking to reduce API usage.
  • Test with a small set of documents first to verify that the reranking behaves as expected.

Security Considerations

  • The API Key is a sensitive credential. Store it in a protected secret vault and never expose it in logs or UI.
  • The Base URL should only be changed if you are sure the endpoint is trustworthy.
  • Ensure that any data sent to NVIDIA’s API complies with your organization’s data‑privacy policies.