Add Opik AutoGen Tracing to example

kedro-agentic-workflows/conf/base/genai-config.yml

@@ -52,12 +52,22 @@ intent_prompt_langfuse:
 #   credentials: opik_credentials
 
 # --- Tracing ---
-intent_tracer_langfuse:
-  type: kedro_datasets_experimental.langfuse.LangfuseTraceDataset
-  credentials: langfuse_credentials
-  mode: langchain    # langchain | openai | sdk
+# intent_tracer_langfuse:
+#  type: kedro_datasets_experimental.langfuse.LangfuseTraceDataset
+#  credentials: langfuse_credentials
+#  mode: langchain    # langchain | openai | sdk | autogen
+
+intent_tracer_opik:
+  type: kedro_datasets_experimental.opik.OpikTraceDataset
+  credentials: opik_credentials
+  mode: langchain    # langchain | openai | sdk | autogen
+
+# autogen_tracer_langfuse:
+#  type: kedro_datasets_experimental.langfuse.LangfuseTraceDataset
+#  credentials: langfuse_credentials
+#  mode: autogen
 
-# intent_tracer_opik:
-#  type: kedro_datasets_experimental.opik.OpikTraceDataset
-#  credentials: opik_credentials
-#  mode: openai    # langchain | openai | sdk
+autogen_tracer_opik:
+  type: kedro_datasets_experimental.opik.OpikTraceDataset
+  credentials: opik_credentials
+  mode: autogen

kedro-agentic-workflows/requirements.txt

@@ -4,7 +4,7 @@ ipython>=8.10
 jupyterlab>=3.0
 notebook
 kedro>=1.2.0
-kedro-datasets[pandas-sqltabledataset,pandas-sqlquerydataset,yaml-yamldataset,opik,langchain-chatopenaidataset, langfuse]~=9.0.0
+kedro-datasets[pandas-sqltabledataset,pandas-sqlquerydataset,yaml-yamldataset,opik,opik-opiktracedataset-autogen,langchain-chatopenaidataset,langfuse]~=9.0.0
 kedro-viz>=12.3.0
 langgraph~=1.0.0
 scikit-learn~=1.7.2

kedro-agentic-workflows/src/kedro_agentic_workflows/pipelines/intent_detection/pipeline.py

@@ -31,7 +31,7 @@ def create_pipeline(**kwargs) -> Pipeline:
                     "params:user_id",
                     "user_data",
                     "session_id",
-                    "intent_tracer_langfuse",
+                    "intent_tracer_opik",
                 ],
                 outputs=["user_context", "session_config"],
                 name="load_context_node",

kedro-agentic-workflows/src/kedro_agentic_workflows/pipelines/response_generation_autogen/nodes.py

@@ -16,6 +16,7 @@ def generate_response(
     intent_detection_result: dict,
     user_context: dict,
     session_config: dict,
+    tracer,
 ) -> dict:
     """
     Run the ResponseGenerationAgent to produce a final answer.
@@ -30,17 +31,37 @@ def generate_response(
         result = {"messages": [AIMessage(content=message)]}
 
     else:
-        agent = ResponseGenerationAgentAutogen(context=response_generation_context)
-        agent.compile()
+        # Wrap agent execution in a span for tracing
+        with tracer.start_as_current_span("response_generation") as span:
+            # Log input context to the span
+            span.set_attribute("intent", intent_detection_result["intent"])
+            span.set_attribute(
+                "intent_reason", intent_detection_result.get("reason", "")
+            )
+            span.set_attribute(
+                "user_id",
+                user_context.get("profile", {}).get("user_id", "unknown"),
+            )
 
-        context = {
-            "messages": [],
-            "intent": intent_detection_result["intent"],
-            "intent_generator_summary": intent_detection_result["reason"],
-            "user_context": user_context,
-        }
+            agent = ResponseGenerationAgentAutogen(context=response_generation_context)
+            agent.compile()
 
-        result = agent.invoke(context, session_config)
+            context = {
+                "messages": [],
+                "intent": intent_detection_result["intent"],
+                "intent_generator_summary": intent_detection_result["reason"],
+                "user_context": user_context,
+            }
+
+            result = agent.invoke(context, session_config)
+
+            # Log output to the span
+            if result.get("messages"):
+                span.set_attribute(
+                    "response", result["messages"][-1].content[:500]
+                )  # Truncate for safety
+            span.set_attribute("claim_created", result.get("claim_created", False))
+            span.set_attribute("escalated", result.get("escalated", False))
 
     for m in result["messages"]:
         try:

kedro-agentic-workflows/src/kedro_agentic_workflows/pipelines/response_generation_autogen/pipeline.py

@@ -1,9 +1,6 @@
 from kedro.pipeline import Pipeline, node, pipeline, llm_context_node, tool
 
-from .nodes import (
-    generate_response,
-    log_response_and_end_session,
-)
+from .nodes import generate_response, log_response_and_end_session
 from .tools import build_lookup_docs, build_get_user_claims, build_create_claim
 
 
@@ -28,6 +25,8 @@ def create_pipeline(**kwargs) -> Pipeline:
                     "intent_detection_result",
                     "user_context",
                     "session_config",
+                    # "autogen_tracer_langfuse"
+                    "autogen_tracer_opik"
                 ],
                 outputs="final_response",
                 name="generate_response_node",
@@ -39,4 +38,4 @@ def create_pipeline(**kwargs) -> Pipeline:
                 name="end_session_node",
             ),
         ]
-    )
+    )