diff --git a/src/glam_extractor/api/hybrid_retriever.py b/src/glam_extractor/api/hybrid_retriever.py
index 771ce0664d..03e087d426 100644
--- a/src/glam_extractor/api/hybrid_retriever.py
+++ b/src/glam_extractor/api/hybrid_retriever.py
@@ -838,11 +838,16 @@ class HybridRetriever:
             "with_payload": True,
         }
         
-        # Add named vector 'using' if in multi-embedding mode
-        if self.use_multi_embedding and using:
-            search_params["using"] = using
-        elif self.use_multi_embedding and self._selected_multi_model:
-            search_params["using"] = self._selected_multi_model.value
+        # Add named vector 'using' ONLY if collection actually has named vectors
+        # Single-vector collections will error with "Not existing vector name" otherwise
+        if self.use_multi_embedding and self.multi_retriever:
+            uses_named = self.multi_retriever.uses_named_vectors(self.collection_name)
+            if uses_named:
+                if using:
+                    search_params["using"] = using
+                elif self._selected_multi_model:
+                    search_params["using"] = self._selected_multi_model.value
+            # else: single-vector collection, don't add 'using' parameter
         
         results = self.qdrant_client.query_points(**search_params)
         
@@ -1139,7 +1144,20 @@ class HybridRetriever:
         """
         from qdrant_client.http import models
         
-        query_vector = self._get_embedding(query, using=using)
+        # Check person collection vector size and use appropriate model
+        person_vector_size = self._get_person_collection_vector_size()
+        person_model = using
+        
+        if person_vector_size == 384 and not using:
+            # Person collection uses MiniLM (384-dim), override model selection
+            person_model = "minilm_384"
+            logger.info(f"Person collection uses 384-dim vectors, using MiniLM model")
+        elif person_vector_size == 1536 and not using:
+            person_model = "openai_1536"
+        elif person_vector_size == 768 and not using:
+            person_model = "bge_768"
+        
+        query_vector = self._get_embedding(query, using=person_model)
         
         try:
             # Build query parameters
@@ -1150,11 +1168,16 @@ class HybridRetriever:
                 "with_payload": True,
             }
             
-            # Add named vector 'using' if in multi-embedding mode
-            if self.use_multi_embedding and using:
-                search_params["using"] = using
-            elif self.use_multi_embedding and self._selected_multi_model:
-                search_params["using"] = self._selected_multi_model.value
+            # Add named vector 'using' ONLY if collection actually has named vectors
+            # Single-vector collections will error with "Not existing vector name" otherwise
+            if self.use_multi_embedding and self.multi_retriever:
+                uses_named = self.multi_retriever.uses_named_vectors("heritage_persons")
+                if uses_named:
+                    if using:
+                        search_params["using"] = using
+                    elif self._selected_multi_model:
+                        search_params["using"] = self._selected_multi_model.value
+                # else: single-vector collection, don't add 'using' parameter
             
             # Add schema-aware filters if provided
             if filter_conditions: