Skip to content

Latest commit

 

History

History
156 lines (108 loc) · 7.13 KB

README.md

File metadata and controls

156 lines (108 loc) · 7.13 KB

Slides Translator Agent

Slides Translator is an AI-powered agent designed to automatically translate the content of a Google Slides presentation into a specified language.

It leverages Large Language Models (LLMs) through the Google AI platform and uses the Google Drive and Google Slides APIs to read the original presentation and create a translated copy.

🏛️ Architecture

Architecture Diagram

The agent is developed locally using the ADK (Agent Development Kit) and its web UI for testing.

Once ready, the agent code is deployed to the Vertex AI Agent Engine in the Google Cloud Platform for production use.

In the production environment:

✨ Features

  • Google Slides & Drive Integration: Securely authenticates with Google services using OAuth2 to access presentations.
  • Automated Presentation Copying: Creates a new, translated version of the presentation in your Google Drive, preserving the original.
  • Content Translation: Extracts and translates all text elements within the slides.
  • Context-Awareness: Allows users to provide additional context (e.g., "this is a technical presentation") to improve translation quality.
  • Agent-based Architecture: Built using the google-adk framework for a modular and robust design.
  • Configurable: Allows setting the underlying LLM models for the agent and translation tasks via environment variables.
  • Concurrent Operations: Utilizes concurrent workers for faster text translation and slide updates.

🖼️ Screenshots

Here are a few screenshots of the Slides Translator Agent in action:

1. ADK Web UI (development):

Web UI

2. Gemini Enterprise (production):

Gemini Enterprise Console

⚙️ Setup & Installation

  1. Clone the repository:

    git clone https://github.com/fmind/slides-translator-agent
cd slides-translator-agent

  2. Install project and hooks: This project uses just as a command runner and uv for package management.

    just install

  3. Create an environment file: Copy the example environment file and fill in the required values.

    cp .env.example .env

    You will need to populate variables in the .env file from your GCO project.

🔑 Authentication

This project requires two layers of authentication:

  1. Google Cloud (for Generative AI): Ensure you have authenticated your environment for Google Cloud access. This typically involves:

    • Installing the Google Cloud CLI (gcloud).
    • Running just cloud-auth to login and update your Application Default Credentials.
    • Ensuring the following APIs are enabled in your Google Cloud project (you can use just cloud-services to enable them):
      • AI Platform API (aiplatform.googleapis.com)
      • Cloud Build API (cloudbuild.googleapis.com)
      • Cloud Run API (run.googleapis.com)

  2. Google Workspace (for Slides & Drive): The agent needs to act on your behalf to read and create presentations. This is done via OAuth 2.0.

    • You must create an OAuth 2.0 Client ID in your Google Cloud project.
    • When creating the credential, configure a redirect URI for the web UI, e.g., http://localhost:8000/dev-ui.
    • Once created, copy the Client ID and Client Secret into the AUTHENTICATION_CLIENT_ID and AUTHENTICATION_CLIENT_SECRET variables in your .env file.
    • Make sure the Google Drive API and Google Slides API are enabled for your project.

OAuth 2.0 Client ID Configuration

🔧 Configuration

The application uses environment variables for configuration (defined in .env):

  • GOOGLE_CLOUD_PROJECT: Your Google Cloud project ID.
  • AUTHENTICATION_CLIENT_ID: Your OAuth 2.0 Client ID.
  • AUTHENTICATION_CLIENT_SECRET: Your OAuth 2.0 Client Secret.
  • MODEL_NAME_AGENT: The agent's LLM model (defaults to gemini-2.5-flash).
  • MODEL_NAME_TRANSLATION: The translation task's LLM model (defaults to gemini-1.5-flash).
  • LOGGING_LEVEL: Sets the logging level (e.g., INFO, DEBUG).
  • GOOGLE_CLOUD_LOCATION: The GCP region for the agent (defaults to global).
  • CONCURRENT_TRANSLATION_WORKERS: Number of concurrent workers for translation (defaults to 10).
  • CONCURRENT_SLIDES_BATCH_UPDATES: Number of slide updates to batch together (defaults to 50).
  • TOKEN_CACHE_KEY: The key used to cache the user's OAuth token in the tool state (defaults to user:slides_translator_token).

🚀 Commands

This project uses just as a command runner. Here are the available commands:

Agent

  • just agent: Run the agent's web UI locally.
  • just agent-web [log_level]: Serve the agent's web UI with a specific log level.
  • just agent-deploy: Deploy the agent to Google Cloud Run and Agent Engine.
  • just agent-requirements: Export agent requirements to requirements.txt.

Cloud

  • just cloud: Run all cloud tasks (authentication and settings).
  • just cloud-auth: Authenticate to Google Cloud.
  • just cloud-services: Enable required Google Cloud services.
  • just cloud-settings: Configure Google Cloud project settings.
  • just cloud-storage: Create a Google Cloud Storage bucket.

Checks

  • just check: Run all checks (lock, code, type, format).
  • just check-code: Check code quality with Ruff.
  • just check-format: Check code formatting with Ruff.
  • just check-lock: Check if uv.lock is up to date.
  • just check-type: Check code types with ty.

Formatting

  • just format: Run all formatters.
  • just format-import: Format imports with Ruff.
  • just format-source: Format source code with Ruff.

Installation

  • just install: Install the project and Git hooks.
  • just install-hooks: Install Git hooks. p* just install-project: Sync project dependencies with uv.

Cleaning

  • just clean: Clean all temporary files.
  • just clean-lock: Remove uv.lock.
  • just clean-python: Remove Python cache files.
  • just clean-ruff: Remove the Ruff cache.
  • just clean-venv: Remove the virtual environment.

📄 License

This project is licensed under the MIT License - see the LICENSE.txt file for details.