Flow Superagent Tutorial: Stock Investment Strategy Advisor

Objective

This tutorial provides an example of using FlowSuperAgent to manage other UtilityAgents. The process logic in FlowSuperAgent is represented as a "Directed Acyclic Graph (DAG)", where each node corresponds to an UtilityAgent, and edges denote the message flow. This structure enables parallel processing, allowing all nodes to begin execution as soon as they receive the necessary input information.

Tutorial Description

Use the AI Refinery SDK to create and run an AI system that can provide suggestions on investing in stocks. In this tutorial, you'll utilize four SearchAgent called "Stock Price Researcher", "Stock Trend Researcher", "Stock Influence Researcher", and "Corporate Finance Researcher" to collect information relevant to user queries in a logical way.

Alongside, three AuthorAgent named "Financial Report Writer", "Public Expectation Writer", and "Investment Suggestion Writer" will assess the gathered data and determine whether it is worth investing in a specific company's stock.

To coordinate the workflow between these agents, a FlowSuperAgent named "Investment Strategy Advisor" will be employed to manage and oversee the entire process.

Tutorial Workflow

The figure below depicts the logical graph flow managed by the "Investment Strategy Advisor", where blue nodes represent instances of SearchAgent, and orange nodes correspond to instances of AuthorAgent.

Example Queries

• Should I invest in Tesla stock this quarter?
• Should I invest in MSFT this quarter?
• Should I invest in NVDA this quarter?

Steps

1. Configuration file

To use the FlowSuperAgent, you need to create a yaml file with all the required configuration. You can customize your assistant agent in this simple YAML configuration file.

As a first step, you need to allow the orchestrator to invoke the given Flow Superagent by listing the superagent in the orchestrator's agent_list.

orchestrator:
  agent_list:
    - agent_name: "Investment Strategy Advisor"

Then, you should define the assistant agents of the FlowSuperAgent as utility agents and list them under utility_agents.

  utility_agents:
    - agent_class: SearchAgent
      agent_name: "Stock Price Researcher"
      agent_description: "Search stock price movements."

    - agent_class: SearchAgent
      agent_name: "Stock Trend Researcher"
      agent_description: "Research stock market trends."

    - agent_class: SearchAgent
      agent_name: "Stock Influence Researcher"
      agent_description: "Search factors influencing stock prices."

    - agent_class: SearchAgent
      agent_name: "Corporate Finance Researcher"
      agent_description: "Research corporate finance and financial reports."

    - agent_class: AuthorAgent
      agent_name: "Public Expectation Writer"
      agent_description: "Summarize public expectations based on stock trends."
      config:
        memory_attribute_key: "investment_analysis"
        title: "Investment Analysis Report"
        leading_questions:
          - question: "How do recent stock market trends reflect public sentiment and expectations for future price movements?"
            prompt: "Analyze the latest stock market trends and explain how they influence public perception of future price changes."
          - question: "What common themes emerge from stock trends that indicate investor confidence or concern?"
            prompt: "Identify key patterns in stock trends that signal investor sentiment. Are there recurring indicators of optimism or fear?"
          - question: "Are there any major economic or geopolitical events influencing public expectations in the market?"
            prompt: "List significant events (economic, political, or industry-specific) that have recently shaped public expectations in stock investments."

    - agent_class: AuthorAgent
      agent_name: "Financial Report Writer"
      agent_description: "Write financial summaries based on corporate finance research."
      config:
        memory_attribute_key: "investment_analysis"
        title: "Investment Analysis Report"
        leading_questions:
          - question: "What are the key takeaways from the latest corporate financial reports, and how do they compare with industry benchmarks?"
            prompt: "Summarize the most important points from the latest financial reports, including performance metrics and comparisons with industry standards."
          - question: "How have recent earnings reports and balance sheet disclosures influenced investor sentiment?"
            prompt: "Explain how recent earnings reports and balance sheets have impacted investor confidence, referencing key financial indicators."
          - question: "Are there any financial indicators or metrics that stand out in the companies under research?"
            prompt: "Highlight notable financial metrics from corporate reports that could be useful in making investment decisions."

    - agent_class: AuthorAgent
      agent_name: "Investment Suggestion Writer"
      agent_description: "Generate insights based on stock research and financial reports."
      config:
        memory_attribute_key: "investment_analysis"
        title: "Investment Analysis Report"
        leading_questions:
          - question: "Which stock or company is being analyzed for investment considerations?"
            prompt: "Identify the stock or company mentioned in the conversation history. Provide context on why it is being analyzed, referencing relevant past discussions or queries."
          - question: "What are the key insights from stock trends, financial reports, and market sentiment?"
            prompt: "Analyze stock price movements, market trends, corporate financial reports, and investor sentiment. Identify significant patterns and factors affecting the stock’s performance."
          - question: "Based on the analysis, should investors consider buying, selling, or holding this stock?"
            prompt: "Evaluate the stock's current valuation, market trends, financial stability, and risk factors. Provide a single, definitive investment decision on whether investors should buy, sell, or hold." 

Then, you should define the FlowSuperAgent and configure its workflow. To do so you should: - Define the vertices as entries in the agent_list of the FlowSuperAgent - Define the edges by specifying for each agent_name in the agent_list the next_step that will be taken.

super_agents:
  - agent_class: FlowSuperAgent
    agent_name: "Investment Strategy Advisor"
    agent_description: "Provides investment insights based on stock and finance research."
    config:
      goal: "Generate investment recommendations based on stock research, trends, financial reports, and public expectations."

      agent_list:
        - agent_name: "Stock Price Researcher"
          next_step:
            - "Stock Trend Researcher"
            - "Stock Influence Researcher"
            - "Financial Report Writer"

        - agent_name: "Stock Trend Researcher"
          next_step:
            - "Public Expectation Writer"

        - agent_name: "Stock Influence Researcher"
          next_step:
            - "Investment Suggestion Writer"

        - agent_name: "Corporate Finance Researcher"
          next_step:
            - "Financial Report Writer"

        - agent_name: "Financial Report Writer"
          next_step:
            - "Investment Suggestion Writer"
            - "Stock Influence Researcher"

        - agent_name: "Public Expectation Writer"
          next_step:
            - "Investment Suggestion Writer"

        - agent_name: "Investment Suggestion Writer"

2. Python file

Now, you can start the development of your assistant using these lines of code:

from dotenv import load_dotenv
from air import login, DistillerClient
import os

load_dotenv() # This loads your ACCOUNT and API_KEY from your local '.env' file

login(
    account=str(os.getenv("ACCOUNT")),
    api_key=str(os.getenv("API_KEY")),
    oauth_server=os.getenv("OAUTH_SERVER", ""),
)

distiller_client = DistillerClient()

project = "stock_invest_advisor"

distiller_client.create_project(
    config_path="config.yaml",
    project=project
)

response = distiller_client.interactive(
    project=project,
    uuid="test_user",
)