feat: updated LLM model versions and made them centrally configurable (#125)

Author: michelle-hadfield-nava

app/src/app_config.py

@@ -63,6 +63,22 @@ def db_session(self) -> db.Session:
     rag_chunk_split_overlap: int = 50
     retrieval_top_k: int = 10
 
+    # LLM Model Configuration per pipeline
+    default_openai_model_version: str = "gpt-5.1"
+    default_openai_reasoning_level: str = "none"
+
+    generate_referrals_model_version: str = "gpt-5.1"
+    generate_referrals_reasoning_level: str = "none"
+
+    generate_referrals_rag_model_version: str = "gpt-5.1"
+    generate_referrals_rag_reasoning_level: str = "none"
+
+    generate_action_plan_model_version: str = "gpt-5.1"
+    generate_action_plan_reasoning_level: str = "none"
+
+    generate_referrals_from_doc_model_version: str = "gpt-5.1"
+    generate_referrals_from_doc_reasoning_level: str = "none"
+
     def chroma_client(self) -> ClientAPI:
         return chromadb.HttpClient(host=self.rag_db_host, port=self.rag_db_port)
 

app/src/common/components.py

@@ -192,8 +192,8 @@ def run(
         self,
         messages: list[ChatMessage],
         domain: str | None = None,
-        model: str = "gpt-5",
-        reasoning_effort: str = "high",
+        model: str = config.default_openai_model_version,
+        reasoning_effort: str = config.default_openai_reasoning_level,
     ) -> dict:
         """
         Run the OpenAI web search generator.

app/src/pipelines/generate_action_plan/pipeline_wrapper.py

@@ -8,6 +8,7 @@
 from opentelemetry.trace.status import Status, StatusCode
 from pydantic import BaseModel
 
+from src.app_config import config
 from src.common import haystack_utils, phoenix_utils
 from src.common.components import (
     LlmOutputValidator,
@@ -98,7 +99,10 @@ def _run(self, resource_objects: list[Resource], user_email: str, user_query: st
                     "action_plan_json": action_plan_as_json,
                     "user_query": user_query,
                 },
-                "llm": {"model": "gpt-5-mini", "reasoning_effort": "low"},
+                "llm": {
+                    "model": config.generate_action_plan_model_version,
+                    "reasoning_effort": config.generate_action_plan_reasoning_level,
+                },
             },
             include_outputs_from={"llm", "save_result"},
         )

app/src/pipelines/generate_referrals/pipeline_wrapper.py

@@ -15,6 +15,7 @@
 from opentelemetry.trace.status import Status, StatusCode
 from pydantic import BaseModel
 
+from src.app_config import config
 from src.common import components, haystack_utils, phoenix_utils
 
 logger = logging.getLogger(__name__)
@@ -154,5 +155,8 @@ def _run_arg_data(
                 "query": query,
                 "response_json": response_schema,
             },
-            "llm": {"model": "gpt-5-mini", "reasoning_effort": "low"},
+            "llm": {
+                "model": config.generate_referrals_model_version,
+                "reasoning_effort": config.generate_referrals_reasoning_level,
+            },
         }

app/src/pipelines/generate_referrals_from_doc/pipeline_wrapper.py

@@ -11,6 +11,7 @@
 from openinference.instrumentation import _tracers, using_metadata
 from opentelemetry.trace.status import Status, StatusCode
 
+from src.app_config import config
 from src.common import components, haystack_utils, phoenix_utils
 from src.pipelines.generate_referrals.pipeline_wrapper import response_schema
 
@@ -95,7 +96,10 @@ def _run(self, files: List[UploadFile]) -> dict:
                 "prompt_builder": {
                     "response_json": response_schema,
                 },
-                "llm": {"model": "gpt-5-mini", "reasoning_effort": "low"},
+                "llm": {
+                    "model": config.generate_referrals_from_doc_model_version,
+                    "reasoning_effort": config.generate_referrals_from_doc_reasoning_level,
+                },
             },
             include_outputs_from={"llm"},
         )

app/src/pipelines/generate_referrals_rag/pipeline_wrapper.py

@@ -84,4 +84,9 @@ def _run_arg_data(
             # For querying RAG DB
             "query_embedder": {"text": query},
             "retriever": {"top_k": config.retrieval_top_k, "filters": None},
+            # Override LLM config for RAG pipeline
+            "llm": {
+                "model": config.generate_referrals_rag_model_version,
+                "reasoning_effort": config.generate_referrals_rag_reasoning_level,
+            },
         }