utils/agentic.py

"""
FinancialInsights.py
A single-file, modular-style agentic workflow for financial news insights.
This file can be imported by a main application (e.g., a FastAPI server)
to run the crew.
"""

import os
from crewai import Agent, Task, Crew, Process, LLM
from crewai_tools import SerperDevTool

# --- 1. Configuration ---

# Using a placeholder for the model name as per your original script.
# Replace with a valid model identifier if needed.
LLM_MODEL = "gemini/gemini-2.5-flash-lite"
LLM_TEMPERATURE = 0.7

SERPER_SEARCH_URL = "https://google.serper.dev/search"
SERPER_N_RESULTS = 5 # Increased results for better financial context

# --- 2. LLM & Tool Initialization ---

# Initialize the LLM instance
llm = LLM(
    model=LLM_MODEL,
    temperature=LLM_TEMPERATURE,
)

# Initialize the search tool
search_tool = SerperDevTool(
    search_url=SERPER_SEARCH_URL,
    n_results=SERPER_N_RESULTS,
)


# --- 3. Agent Definitions ---

def create_financial_researcher():
    """Creates the Financial Researcher agent."""
    return Agent(
        role="Financial Researcher",
        goal="Gather the latest news, market sentiment, and key developments "
             "for the given list of financial topics, companies, and cryptocurrencies. "
             "Use the Serper tool to find relevant articles, reports, and discussions.",
        backstory=(
            "You are an expert financial researcher, skilled at digging through "
            "market news, press releases, and financial forums to find the most "
            "relevant and timely information for an investor."
        ),
        tools=[search_tool],
        verbose=True,
        memory=True,
        llm=llm,
    )

def create_financial_analyst():
    """Creates the Financial Analyst agent."""
    return Agent(
        role="Financial Analyst",
        goal="Analyze the raw research findings and synthesize them into a "
             "consolidated, easy-to-read investor insight report. "
             "Identify key trends, potential risks, and opportunities. "
             "The final output must be a professional report for an investor.",
        backstory=(
            "You are a seasoned financial analyst with a talent for "
            "cutting through the noise. You can take a pile of raw data and "
            "news snippets and distill them into actionable insights and "
            "concise summaries for busy investors."
        ),
        tools=[],  # No external tools needed; analyzes data from the researcher
        verbose=True,
        memory=True,
        allow_delegation=False,
        llm=llm,
    )


# --- 4. Task Definitions ---

def create_research_task(agent):
    """Creates the research task for the given agent."""
    return Task(
        description=(
            "For each topic in the provided list {interests}, search for the "
            "latest news, market analysis, and significant events from the past 48 hours. "
            "Gather key snippets, sources, and general sentiment."
        ),
        expected_output=(
            "A structured report with sections for each topic, each containing:\n"
            "Topic: <topic name>\n"
            "Key News Snippets:\n- [Source]: <snippet>...\n"
            "Market Sentiment:\n- <Summary of general sentiment (e.g., bullish, bearish, neutral)>\n"
            "Recent Developments:\n- <Bulleted list of key events or changes>"
        ),
        agent=agent,
        tools=[search_tool],
    )

def create_analysis_task(agent):
    """Creates the analysis and reporting task for the given agent."""
    return Task(
        description=(
            "Take the researcher's raw findings for all {interests}. "
            "Analyze and synthesize this information into a consolidated 'Investor Insight' report. "
            "Start with a high-level executive summary, then provide a detailed "
            "breakdown for each topic. Focus on what this information *means* "
            "for an investor, highlighting key insights, risks, and potential opportunities."
        ),
        expected_output=(
            "A comprehensive investor report in markdown format.\n\n"
            "## Executive Summary\n"
            "<Brief overview of all topics and major market movements.>\n\n"
            "## Detailed Insights\n\n"
            "### [Topic 1 Name]\n"
            "- **Key Insight:** <What is the most important takeaway?>\n"
            "- **Recent News:** <Summary of the most impactful news.>\n"
            "- **Potential Risk:** <Identify a potential risk.>\n"
            "- **Potential Opportunity:** <Identify a potential opportunity.>\n\n"
            "### [Topic 2 Name]\n"
            "- **Key Insight:** <...>\n"
            "- **Recent News:** <...>\n"
            "- **Potential Risk:** <...>\n"
            "- **Potential Opportunity:** <...>\n"
            "(...and so on for all topics)"
        ),
        agent=agent,
    )


# --- 5. Crew Definition & Workflow Function ---

def run_financial_crew(interest_list):
    """
    Initializes and kicks off the financial insights crew with a given list of interests.
    This is the main function to be imported and called by an external app.
    
    Args:
        interest_list (list): A list of financial topics (strings) to research.
        
    Returns:
        str: The final result from the crew's execution (the investor report).
    """
    # 1. Create Agents
    financial_researcher = create_financial_researcher()
    financial_analyst = create_financial_analyst()
    
    # 2. Create Tasks
    research_task = create_research_task(financial_researcher)
    analysis_task = create_analysis_task(financial_analyst)
    
    # 3. Create Crew
    financial_crew = Crew(
        agents=[financial_researcher, financial_analyst],
        tasks=[research_task, analysis_task],
        process=Process.sequential, # Sequential: Researcher -> Analyst
    )
    
    # 4. Prepare Inputs
    inputs = {"interests": interest_list}
    
    # 5. Run Crew
    print(f"Starting crew for financial interests: {interest_list}...")
    result = financial_crew.kickoff(inputs=inputs)
    return result


# --- 6. Independent Run (for testing) ---

if __name__ == "__main__":
    """
    This block allows the script to be run directly for testing purposes.
    It will not execute when the file is imported as a module.
    """
    
    # Define the list of financial interests to research
    interests = ["NVIDIA stock", "US inflation rate"]
    
    # Run the crew workflow
    result = run_financial_crew(interests)
    
    # Print the final output
    print("\n=== Final Investor Insight Report ===\n")
    print(result)