diff --git a/agenta-backend/agenta_backend/models/api/evaluation_model.py b/agenta-backend/agenta_backend/models/api/evaluation_model.py
index 0f2b1b364a..913f00c500 100644
--- a/agenta-backend/agenta_backend/models/api/evaluation_model.py
+++ b/agenta-backend/agenta_backend/models/api/evaluation_model.py
@@ -4,6 +4,7 @@
 
 from pydantic import BaseModel, Field, model_validator
 
+from agenta_backend.utils import traces
 from agenta_backend.models.api.api_models import Result
 
 
@@ -98,6 +99,15 @@ class EvaluatorMappingInputInterface(BaseModel):
     inputs: Dict[str, Any]
     mapping: Dict[str, Any]
 
+    @model_validator(mode="before")
+    def remove_trace_prefix(cls, values: Dict) -> Dict:
+        mapping = values.get("mapping", {})
+        updated_mapping = traces.remove_trace_prefix(mapping_dict=mapping)
+
+        # Set the modified mapping back to the values
+        values["mapping"] = updated_mapping
+        return values
+
 
 class EvaluatorMappingOutputInterface(BaseModel):
     outputs: Dict[str, Any]
diff --git a/agenta-backend/agenta_backend/services/evaluators_service.py b/agenta-backend/agenta_backend/services/evaluators_service.py
index 3e5b1cd6f6..ae8bad50e9 100644
--- a/agenta-backend/agenta_backend/services/evaluators_service.py
+++ b/agenta-backend/agenta_backend/services/evaluators_service.py
@@ -21,6 +21,7 @@
     EvaluatorMappingOutputInterface,
 )
 from agenta_backend.utils.traces import (
+    remove_trace_prefix,
     process_distributed_trace_into_trace_tree,
     get_field_value_from_trace_tree,
 )
@@ -934,9 +935,10 @@ async def rag_faithfulness(
             )
 
         # Get required keys for rag evaluator
-        question_key: Union[str, None] = settings_values.get("question_key", None)
-        answer_key: Union[str, None] = settings_values.get("answer_key", None)
-        contexts_key: Union[str, None] = settings_values.get("contexts_key", None)
+        mapping_keys = remove_trace_prefix(settings_values=settings_values)
+        question_key: Union[str, None] = mapping_keys.get("question_key", None)
+        answer_key: Union[str, None] = mapping_keys.get("answer_key", None)
+        contexts_key: Union[str, None] = mapping_keys.get("contexts_key", None)
 
         if None in [question_key, answer_key, contexts_key]:
             logging.error(
@@ -1046,9 +1048,10 @@ async def rag_context_relevancy(
             )
 
         # Get required keys for rag evaluator
-        question_key: Union[str, None] = settings_values.get("question_key", None)
-        answer_key: Union[str, None] = settings_values.get("answer_key", None)
-        contexts_key: Union[str, None] = settings_values.get("contexts_key", None)
+        mapping_keys = remove_trace_prefix(settings_values=settings_values)
+        question_key: Union[str, None] = mapping_keys.get("question_key", None)
+        answer_key: Union[str, None] = mapping_keys.get("answer_key", None)
+        contexts_key: Union[str, None] = mapping_keys.get("contexts_key", None)
 
         if None in [question_key, answer_key, contexts_key]:
             logging.error(
diff --git a/agenta-backend/agenta_backend/services/llm_apps_service.py b/agenta-backend/agenta_backend/services/llm_apps_service.py
index 0cdef1a2d2..d87177ff6e 100644
--- a/agenta-backend/agenta_backend/services/llm_apps_service.py
+++ b/agenta-backend/agenta_backend/services/llm_apps_service.py
@@ -55,13 +55,10 @@ def extract_result_from_response(response: dict):
                 value["data"] = str(value.get("data"))
 
             if "tree" in response:
-                trace_tree = (
-                    response["tree"][0]
-                    if isinstance(response.get("tree"), list)
-                    else {}
-                )
+                trace_tree = response.get("tree", {}).get("nodes", [])[0]
+
                 latency = (
-                    get_nested_value(trace_tree, ["time", "span"]) * 1_000_000
+                    get_nested_value(trace_tree, ["time", "span"]) / 1_000_000
                     if trace_tree
                     else None
                 )
@@ -108,6 +105,8 @@ def extract_result_from_response(response: dict):
         value = {"error": f"Unexpected error: {e}"}
         kind = "error"
 
+    print("Cost: ", cost)
+    print("Latency: ", latency)
     return value, kind, cost, latency