Tutorial: Building a multi-modal researcher with Gemini 2.5

Full tutorial on YouTube

Code on GitHub

Note on Notion

Tutorial summary:

Lance from Langchain demonstrates a simple multi-modal researcher built with Gemini 2.5 models. This tool uses Gemini’s native search, YouTube video understanding, and text-to-speech capabilities to research a topic and a related video, combine the findings into a report, and generate a podcast summarizing the results. The workflow is orchestrated in Langraph Studio, allowing users to input a topic and YouTube URL, then view each step’s inputs and outputs. The Gemini 2.5 models are highlighted for their strong performance, multimodal abilities, and ease of integration. The code is modular, with nodes for search, video analysis, report generation, and podcast creation. Users can easily configure model parameters in the Studio interface. The project is open source and customizable, encouraging experimentation with Gemini’s advanced features.

Tips & Tricks for Running This App on Windows with PowerShell

1. Installing uvx with pipx

• Open PowerShell and install pipx (if not already installed):

  pip install pipx
  

• Install uv (which provides uvx):

  pipx install uv
  

• If you see a warning about C:\Users\<your-username>\.local\bin not being in your PATH, run:

  pipx ensurepath
  

  Then restart PowerShell to update your PATH.

2. Making Sure uvx is Available

• After installation, check if uvx is available:

  uvx --version
  

• If you get a 'uvx' is not recognized error, either:
  • Restart PowerShell (to reload your PATH), or
  • Use the full path to run uvx:

    & "$env:USERPROFILE\.local\bin\uvx.exe" --version
    

3. Setting Environment Variables

Option A: Use a .env File (Recommended)

• Copy .env.example to .env:

  cp .env.example .env
  

• Edit .env and add your API keys:

  GEMINI_API_KEY=your_gemini_api_key
  LANGSMITH_API_KEY=your_langsmith_api_key
  

• The app will automatically load variables from .env (as specified in langgraph.json).

Option B: Set Variables in PowerShell Session

• Set variables for the current session:

  $env:GEMINI_API_KEY="your_gemini_api_key"
  $env:LANGSMITH_API_KEY="your_langsmith_api_key"
  

• Important: You must start the app from the same PowerShell window.

Option C: Set System Environment Variables

• Add variables via Windows System Properties > Environment Variables.
• Note: You must log out and log back in (or restart) for changes to take effect in new terminals.

4. Running the App

• Use the following command to install dependencies and start the server:

  uvx --refresh --from "langgraph-cli[inmem]" --with-editable . --python 3.11 langgraph dev --allow-blocking
  

• If uvx is not recognized, use the full path as shown above.

5. Accessing the Application

• After starting, the app will print URLs such as:
  • API: http://127.0.0.1:2024
  • Studio UI: https://smith.langchain.com/studio/?baseUrl=http://127.0.0.1:2024
  • API Docs: http://127.0.0.1:2024/docs

6. Common Issues & Solutions

Summary:
Carefully follow the installation and environment setup steps above to ensure a smooth experience running this app on Windows with PowerShell. Most issues are related to environment variables or PATH settings—restart your terminal or computer after making changes!