Design the LlamaIndex Financial Agent and Tools

Design a LlamaIndex agent named 'FinancialAnalystAgent' that takes structured financial data (e.g., CSV content) and projection parameters as input. Define custom tools for this agent, such as `parse_financial_csv`, `calculate_projection`, and `generate_audit_step`. These tools will use GPT-5 for complex reasoning within `calculate_projection` and Claude Sonnet 4 for generating concise `generate_audit_step` summaries. Describe how these tools would be integrated into the LlamaIndex agent's query engine. ```python
from llama_index.core.agent import FunctionCallingAgentWorker, AgentRunner
from llama_index.llms.openai import OpenAI # Assuming GPT-5 via OpenAI API
from llama_index.llms.anthropic import Anthropic # Assuming Claude Sonnet 4 via Anthropic API
from llama_index.core.tools import FunctionTool # Define tool functions here
def parse_financial_csv(csv_content: str) -> dict: # Implementation to parse CSV into structured data return {"data": []} def calculate_projection(data: dict, params: dict) -> dict: # Use GPT-5 for complex calculation logic llm_gpt5 = OpenAI(model="gpt-5") # ... logic here ... return {"projection": {}} def generate_audit_step(action_description: str) -> str: # Use Claude Sonnet 4 for audit trail generation llm_claude_sonnet_4 = Anthropic(model="claude-sonnet-4") # ... logic here ... return f"Audit: {action_description}" # Instantiate tools
parse_tool = FunctionTool.from_defaults(fn=parse_financial_csv, name="parse_financial_csv", description="...")
# ... other tool instantiations ... # Create agent worker and runner
# worker = FunctionCallingAgentWorker.from_tools([parse_tool, ...], llm=llm_gpt5)
# agent = AgentRunner(worker)
```