Lab guide for financial research and analysis¶

Overview¶

A financial advisor must go through various data points before meeting a client for financial advisory. The financial advisor must spend time after the meeting with the client to summarize the key decisions that were suggested during the meeting. The financial research and analysis lab focusses on various Gen AI use cases such as retrieval augmented generation, SQL retrieval, web search, summarization and text generation that will help boost the productivity of the financial advisor.

Pre-requisites¶

• Make sure you've already setup the environment:
• Lab 0 - Environment setup
• ADK Installation
• Download files
• Download the financial_research_and_analysis_agent.zip file from Lab2 folder.

Reference Architecture¶

Key Components¶

• Risk assessment agent (External LangGraph Agent) – An agent running in a custom langchain container capable of communicating with external news sources to combat disinformation.
• Report generator agent (Collaborator Agent) – An agent capable of generating stock investment portfolio reports by retrieving user portfolios and summarizing relevant articles.
• Financial research and analysis agent (Supervisor Agent) - An agent capable of providing financial insights, generating stock investment reports, and assisting with meeting follow-ups.
• Portfolio retriever (Tool) – Get client's portfolio from database.
• Transcript retriever (Tool) – Get client's meeting transcripts from database.
• Web search (Tool) - Search the web for relevant information.
• Goldman Sachs 2025 outlook (Knowledge base) - This knowledge addresses the goldman sachs 2025 outlook.

Steps¶

Make sure you have downloaded the required files for the lab and extracted it. Go to the root of the downloaded directory.

cd financial_research_and_analysis_agent/

You should see the following directories & files listed:

├── agents
│   ├── agent.yaml
│   └── report_generator_agent.yaml
├── cleanup.sh
├── collaborators
│   ├── external-agent-source-code.zip
│   └── external-agent.yaml
├── knowledge_bases
│   ├── 2025Outlook.pdf
│   └── knowledge.yaml
├── README.md
├── setup.sh
└── tools
    ├── portfolio_retriever_tool
    │   ├── portfolio_retriever.py
    │   └── requirements.txt
    ├── transcript_retriever_tool
    │   ├── requirements.txt
    │   └── transcript_retriever.py
    └── web_search_tool
        ├── requirements.txt
        └── web_search.py

8 directories, 15 files

1. Deploy the tools¶

Portfolio retriever tool & Transcript retriever tool¶

• Deploy the portfolio_retriever_tool and transcript_retriever_tool directly by running the following commands:

  orchestrate tools import -k python -f "tools/portfolio_retriever_tool/portfolio_retriever.py" -r "tools/portfolio_retriever_tool/requirements.txt"
  

  orchestrate tools import -k python -f "tools/transcript_retriever_tool/transcript_retriever.py" -r "tools/transcript_retriever_tool/requirements.txt"
  

Web search tool¶

We have integrated the Tavily search API to implement the web search tool within our watsonx Orchestrate. Tavily is a third-party AI-powered search service designed to deliver accurate, real-time information by querying the web and returning concise, relevant results. This integration demonstrates how external APIs can be seamlessly incorporated into watsonx Orchestrate to enhance agent capabilities with dynamic and up-to-date knowledge retrieval from the internet.

• Create a connection with watsonx Orchestrate.

  orchestrate connections add -a tavily
  

  orchestrate connections configure --app-id tavily --environment draft -t team -k key_value
  

  orchestrate connections set-credentials --app-id=tavily --env draft -e TAVILY_API_KEY=<your_api_key>
  

• Deploy the tool.

  orchestrate tools import -k python -f "tools/web_search_tool/web_search.py" -r "tools/web_search_tool/requirements.txt" --app-id tavily
  

At this point you have successfully deployed all the required tools.

2. Deploy the knowledge base¶

We are using a subset of Goldman Sach's 2025 outlook report as a knowledge for the financial research and analysis agent. Run the following command to deploy the knowledge base:

orchestrate knowledge-bases import -f knowledge_bases/knowledge.yaml

At this point you have successfully deployed all the knowledge bases.

3. Deploy the external agent¶

We have developed a Risk Assessment Agent using LangGraph and integrated it as an external collaborator within watsonx Orchestrate. This implementation demonstrates how open-source agent frameworks can interoperate seamlessly with watsonx Orchestrate native agents, enabling flexible, modular, and extensible agent collaboration across platforms.

Now that you have obtained the deployed api url from instructor and patched the external_agent.yaml file, navigate back to the root directory financial_research_and_analysis_agent and run the following commands to deploy the agent on watsonx Orchestrate.

orchestrate agents import -f collaborators/external_agent.yaml

At this point, the external agents are deployed to watsonx Orchestrate.

4. Deploy supervisor and collaborative agents¶

Finally, deploy the Financial research and analysis agent and the Report generator agent by running the following commands:

orchestrate agents import -f agents/report_generator_agent.yaml

orchestrate agents import -f agents/agent.yaml

At this point, all the required agents and tools are deployed on watsonx Orchestrate and ready to use.

Now you can chat with the agent with the suggested script below.

Suggested script¶

Select the financial_research_and_analysis_agent.

1. Generate a portfolio report on John Doe's securities -> The financial research and analysis agent will perform the following tasks to get the report:

   • Get the portfolio of John Doe from a SQL database.
   • Transfer the control to the Report generator agent.
   • The report generator agent will search each of the security performance on web using web search tool.

   • The report generator agent will summarize everything and generate a report.

     

2. Perform a risk assessment on the securities -> The financial research and analysis agent will perform the following tasks to get the risk assessment report:

   • Transfer the control to the external risk assessment agent with the securities.

   • The external risk assessment agent will perform the risk assessment with its business logic and return the report.

     

     Bug

     Currently the transfer back from external agent to supervisor is not happening, this is a known bug and the product team is working on fixing it!

     • Kindly start a new chat after asking the above question.

3. Summarize recent meeting with John Doe and generate an email with minutes of meeting -> The financial research and analysis agent will perform the following tasks to generate email:

   • Get the transcript of John Doe's meeting from Object Storage.

   • Summarize it and create an email.

     

4. How have interest rate cuts impacted valuations according to Goldman Sachs Report? or What is Goldman’s outlook on global diversification? -> The financial research and analysis agent will retrieve the information from the Goldman Sachs 2025 outlook report (RAG) to answer these queries.

   

5. How does the US equity market compare to the rest of world? -> The financial research and analysis agent will perform the following tasks to give the financial insights:

   • Use the Tavily search to get the results

     Note: If you have not set the Tavily API key then this won't work