Integrate GPT-5 Pro and Gemini 3 Flash via AI21 Studio/Together AI

Implement the 'Bias Analyzer' node using GPT-5 Pro and the 'Fact Checker' node using Gemini 3 Flash within your LangGraph setup. For GPT-5 Pro, focus on its ability to identify subtle linguistic cues, rhetorical strategies, and omitted information that indicate bias. For Gemini 3 Flash, create a `Tool` that it can use to query a mock external fact database (e.g., a dictionary of verified claims or a simple API for public data). Utilize AI21 Studio for deploying GPT-5 Pro inference and Together AI for serving Gemini 3 Flash models, ensuring high performance and availability. Ensure API keys and endpoints are configured securely for both models.
```python
import os
from langchain_openai import ChatOpenAI # For GPT-5 Pro (assuming OpenAI-compatible API endpoint like AI21 Studio's)
from langchain_google_genai import ChatGoogleGenerativeAI # For Gemini 3 Flash
from langchain_core.tools import tool # --- GPT-5 Pro via AI21 Studio --- (Conceptual integration)
# AI21 Studio's models might have their own specific LangChain integration or be OpenAI-compatible.
# For GPT-5 Pro, we'll assume an OpenAI-compatible interface through AI21 Studio.
# gpt5_pro_llm = ChatOpenAI(
# model="gpt-5-pro", # Placeholder, use actual model ID from AI21 Studio
# openai_api_base=os.getenv("AI21_STUDIO_API_BASE_URL", "https://api.ai21.com/v1"), # Replace with AI21 Studio's endpoint
# api_key=os.getenv("AI21_STUDIO_API_KEY"), # Your AI21 Studio API key
# temperature=0.3
# ) # def bias_analyzer_node(state: AgentState) -> AgentState:
# # Logic for Bias Analyzer using gpt5_pro_llm
# analysis_prompt = f"Analyze the following text for bias and provide reasoning: {state['content_to_review']}"
# response = gpt5_pro_llm.invoke(analysis_prompt)
# state['bias_analysis'] = response.content
# state['has_issues'] = "biased" in response.content.lower() # Simple check
# return state # --- Gemini 3 Flash via Together AI --- (Conceptual integration)
# gemini3_flash_llm = ChatGoogleGenerativeAI(
# model="gemini-3-flash", # Model name as provided by Together AI for Gemini 3 Flash
# base_url=os.getenv("TOGETHER_AI_API_BASE_URL", "https://api.together.ai/v1"), # Together AI endpoint
# api_key=os.getenv("TOGETHER_AI_API_KEY"), # Your Together AI API key
# temperature=0.1
# ) # Define a custom tool for fact-checking with Gemini 3 Flash
# @tool
# def search_fact_database(query: str) -> str:
# """Searches a mock fact database for information related to the query. Returns 'Verified' or 'Unverified' with context."""
# # Simulate a database lookup
# if "economy will be destroyed" in query.lower():
# return "Unverified. Economic forecasts vary; no consensus on 'destruction'."
# return "Verified. General statement." # Mock response # def fact_checker_node(state: AgentState) -> AgentState:
# # Logic for Fact Checker using gemini3_flash_llm and tools
# claims_to_check = extract_claims_from_content(state['content_to_review']) # Custom function
# checked_facts = []
# for claim in claims_to_check:
# tool_response = search_fact_database.invoke({'query': claim})
# checked_facts.append({"claim": claim, "status": tool_response})
# state['fact_check_results'] = checked_facts
# state['has_issues'] = state['has_issues'] or any("Unverified" in f['status'] for f in checked_facts)
# return state # Add these functions as nodes to your graph_builder as defined in Prompt 1
# graph_builder.add_node("bias_analyzer", bias_analyzer_node)
# graph_builder.add_node("fact_checker", fact_checker_node)
```