rerankSearchResults

Takes a list of vector search results and a query, then reorders them by relevance using either a dedicated reranker model (RunPod or Replicate) or an LLM-based scoring fallback. Returns the top-K results with both original similarity scores and new rerank scores.

Execution path:

1. If use_dedicated_reranker is true and provider is not "llm", calls a dedicated reranker (RunPod / Replicate). Falls through to the LLM path on failure.
2. Otherwise, computes relevance scores via qwen3-235b-a22b-instruct-2507 on Scaleway AI.
3. If the LLM response cannot be parsed, the original similarity scores are used as-is.

An empty search_results input returns immediately with count: 0.

Parameters

Param	Type	Required	Default	Description
search_results	SearchResult[]	Yes	—	List of results to rerank. Each item must have page_content, metadata, and score
query	string	Yes	—	Original search query. Minimum length: 1
top_k	integer	No	5	Number of top results to return after reranking
rerank_model	string	No	"BAAI/bge-reranker-large"	Model to use when using a dedicated reranker
provider	string?	No	null	Force a specific provider: runpod, replicate, or llm. Auto-detected from model when null
use_dedicated_reranker	boolean	No	true	When true, attempts dedicated reranker before falling back to LLM scoring

Output

Field	Type	Description
ranked_results	RankedResult[]	Reranked results, ordered by rerank_score descending, truncated to top_k
count	integer	Number of results returned
provider	string	Provider used: runpod, replicate, llm, none (empty input), or error
model	string	Model used for reranking

Each RankedResult:

Field	Type	Description
page_content	string	Text content of the chunk
metadata	object	Document metadata
similarity_score	float	Original similarity score from the vector search
rerank_score	float	Score assigned by the reranker
final_rank	integer	1-based position in the final ordered list

Example

{
  "id": "rerank",
  "type": "rerankSearchResults",
  "data": {
    "label": "Rerank Results",
    "isExecuted": false,
    "handles": ["inputs", "outputs"],
    "schema": {},
    "params": {
      "search_results": { "value": "{{ @vectorSearch.documents }}", "isExpression": true, "isAttachedToInputNode": false },
      "query": { "value": "{{ $input.query }}", "isExpression": true, "isAttachedToInputNode": false },
      "top_k": { "value": 5, "isExpression": false, "isAttachedToInputNode": false },
      "use_dedicated_reranker": { "value": true, "isExpression": false, "isAttachedToInputNode": false }
    },
    "inputs": [], "outputs": [], "errors": []
  },
  "position": { "x": 300, "y": 0 },
  "isSelected": false,
  "isDragging": false
}