Reranking nodes use Large Language Models to improve the quality and relevance of retrieved documents in your RAG pipeline by intelligently reordering results.

​
What are Reranking Nodes?

Reranking nodes are processing components that use LLMs (like GPT-4) to evaluate and reorder documents retrieved by your system. They analyze documents with deep semantic understanding to improve result relevance.

​
Key Features

  • LLM-Powered: Use advanced language models for intelligent document scoring
  • Quality Enhancement: Improve result precision and user satisfaction
  • Flexible Configuration: Adjust the number of results to return (Top K)
  • Performance Control: Balance quality improvements with processing speed

​
Available Endpoints

EndpointMethodPurpose
List Reranking NodesGETGet reranking node configurations
Update Reranking ConfigurationPATCHModify reranking settings

​
Configuration

​
Top K Parameter

The main configuration option is topK - the maximum number of documents to return after reranking:
  • Small values (3-5): Fast processing, highest precision
  • Medium values (8-15): Balanced performance and quality
  • Large values (20-30): More comprehensive results, slower processing
  • null: Process all documents (highest resource usage)

​
Example Configuration

{
  "config": {
    "topK": 10
  }
}

​
Node Structure

{
  "id": "reranking-1748287628687",
  "type": "reranking",
  "data": {
    "name": "LLM Reranking",
    "config": {
      "topK": 10
    },
    "result": {
      "updated": true,
      "processing": false,
      "total_reranked": 847
    }
  }
}

​
Common Use Cases

​
Customer Support

  • Top K: 5-8
  • Goal: Provide precise answers to customer queries
  • Benefits: Faster resolution, better satisfaction
  • Top K: 10-15
  • Goal: Help users find information efficiently
  • Benefits: Better search experience

​
Research Applications

  • Top K: 20-30 or null
  • Goal: Comprehensive results for analysis
  • Benefits: Thorough coverage, better insights

​
Authentication

All endpoints require API token authentication: 
Authorization: Bearer YOUR_API_TOKEN

​
URL Structure

https://{flow_name}.flows.graphorlm.com/reranking[/{node_id}]

​
Basic Usage Example

// List all reranking nodes
const response = await fetch('https://my-flow.flows.graphorlm.com/reranking', {
  headers: { 'Authorization': 'Bearer YOUR_API_TOKEN' }
});
const nodes = await response.json();

// Update a reranking node
await fetch('https://my-flow.flows.graphorlm.com/reranking/node-id', {
  method: 'PATCH',
  headers: {
    'Authorization': 'Bearer YOUR_API_TOKEN',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    config: { topK: 15 }
  })
});

​
Best Practices

  1. Start Small: Begin with lower Top K values and increase as needed
  2. Monitor Performance: Track processing time and resource usage
  3. Test Changes: Compare results before and after configuration changes
  4. Match Use Case: Choose Top K based on your specific requirements

​
Error Handling

Common errors and solutions:
  • “topK must be a positive integer or null”: Check parameter type
  • “Reranking node not found”: Verify node ID
  • “Failed to update reranking node”: Check system status and retry

​
Next Steps

List Reranking Nodes

View your current reranking configurations

Update Configuration

Optimize your reranking settings

API Tokens

Learn about API authentication

Flow Management

Explore other flow endpoints