feat(config): Implement asymmetric model configuration for Orchestrator and Workers

backend/src/ea_chatbot/config.py

@@ -52,6 +52,8 @@ class Settings(BaseSettings):
     coder_llm: LLMConfig = Field(default_factory=lambda: LLMConfig(model="gpt-5-mini", temperature=0.0))
     summarizer_llm: LLMConfig = Field(default_factory=lambda: LLMConfig(model="gpt-5-mini", temperature=0.0))
     researcher_llm: LLMConfig = Field(default_factory=lambda: LLMConfig(model="gpt-5-mini", temperature=0.0))
+    reflector_llm: LLMConfig = Field(default_factory=lambda: LLMConfig(model="gpt-5-mini", temperature=0.0))
+    synthesizer_llm: LLMConfig = Field(default_factory=lambda: LLMConfig(model="gpt-5-mini", temperature=0.0))
     
     # Allow nested env vars like QUERY_ANALYZER_LLM__MODEL
     model_config = SettingsConfigDict(env_nested_delimiter='__', env_prefix='')

backend/src/ea_chatbot/graph/nodes/reflector.py

@@ -22,7 +22,7 @@ def reflector_node(state: AgentState) -> dict:
     logger.info(f"Evaluating worker output for task: {task_desc[:50]}...")
     
     llm = get_llm_model(
-        settings.planner_llm, # Using planner model for evaluation
+        settings.reflector_llm,
         callbacks=[LangChainLoggingHandler(logger=logger)]
     )
     structured_llm = llm.with_structured_output(ReflectorResponse)

backend/src/ea_chatbot/graph/nodes/synthesizer.py

@@ -19,7 +19,7 @@ def synthesizer_node(state: AgentState) -> dict:
     logger.info("Synthesizing final answer from worker results...")
     
     llm = get_llm_model(
-        settings.summarizer_llm,
+        settings.synthesizer_llm,
         callbacks=[LangChainLoggingHandler(logger=logger)]
     )