Turn your email newsletters into a daily podcast!

Albert is a friendly AI assistant that lives in your browser. He reads your inbox, understands your intent (e.g., "AI news", "Project updates"), and uses Vertex AI Text-to-Speech to turn it into an engaging audio podcast.

Powered by Google Agent Development Kit (ADK) and Gemini 2.5 Flash, Albert uses a sophisticated multi-agent loop to ensure your digest is high-quality, witty, and concise.

• Semantic Search: Just say "What's happening in AI?" or "Summarize my project updates". Albert uses Gemini Embeddings to find relevant emails even without explicit labels.
• Editorial Quality: A dedicated Refinement Loop (Draft -> Critique -> Refine) ensures the digest sounds like a professional podcast (think NYT Hardfork style).
• Smart Memory: Albert remembers your last query. Just say "Run it again" to get your usual fix.
• Audio Magic: Automatically uses Vertex AI Text-to-Speech to generate realistic, conversational audio mimic NYT Hardfork podcast style.
• Fun Interface: A playful, KAWS-inspired design that makes news less boring.

Albert is a personal concierge assistant built on top of Google Agent Development Kit (ADK) framework, employing a multi-agent loop powered by Gemini 2.5 Flash and Vertex AI Text-to-Speech services to deliver high-quality, witty, and concise daily digests for users on the go.

1. Concierge Agent (Orchestrator): * Manages the user session and coordinates the workflow.
   * Deterministic Handoff: Once the content is ready, the Concierge directly calls the TTS service, ensuring 100% reliability.
   

2. Content Generation Pipeline (Google ADK): * Email Aggregator: Fetches up to 50 emails based on user intent and semantic search. * Refinement Loop (LoopAgent): * Drafter: Synthesizes the initial digest in the style of a news editor.
   * Critic: Reviews the digest for quality, tone, and conciseness.
   * Loop continues until the Critic approves or max iterations reached.
   

3. Audio Generation (Service Layer): * Goal: Convert the finalized text digest into an audio format ("podcast").
   * Tool: Google Cloud Vertex AI Text-to-Speech.
   * Process:
   1. The Concierge receives the approved text from the ADK pipeline.
   2. It calls the TextToSpeechService directly.
   3. The audio file is saved locally and a playback link is returned to the user.
   

graph TD

  %% Client
  U[User]
  UI[Albert Web UI]
  U --> UI

  %% Backend
  subgraph Backend
    CONC[Concierge Agent]
    EA[Email Aggregator]
    D[Drafter]
    C[Critic]
    TTSWRAP[Text-to-Speech Wrapper]
  end

  UI -->|HTTP/API| CONC
  CONC --> EA

  %% Email aggregation
  EA -->|Fetch emails| GM[Gmail API]

  %% Content pipeline
  EA --> D
  D --> C
  C --> D
  C -->|Approved digest| CONC

  %% Vertex AI
  subgraph VertexAI
    LLM[Gemini 2.5 Flash]
    TTS[Vertex AI TTS]
  end

  D --> LLM
  C --> LLM
  LLM --> D
  LLM --> C

  %% TTS + storage
  CONC -->|Final text| TTSWRAP
  TTSWRAP --> TTS
  TTS -->|Audio| GCS[Cloud Storage]
  GCS -->|Playback URL| CONC
  CONC --> UI

  %% Observability
  subgraph Observability
    LOG[Cloud Logging]
    TRACE[Cloud Trace]
  end

  CONC --> LOG
  EA --> LOG
  D --> LOG
  C --> LOG
  TTSWRAP --> LOG

  CONC --> TRACE
  EA --> TRACE
  D --> TRACE
  C --> TRACE
  TTSWRAP --> TRACE

Albert is designed with enterprise-grade observability to help you debug agent behavior and optimize performance.

Visualize the full latency breakdown of your agent pipeline.

• Trace Spans: See exactly how long each step takes (e.g., process_request, generate_audio, adk_pipeline).
• Attributes: Each trace includes metadata like session_id, model, and user_input.

Structured logs for deep debugging.

• Agent Thoughts: View the raw output from the Drafter and Critic agents.
• Tool Calls: See exactly which tools were called and with what arguments.
• Errors: Stack traces and error messages are captured with full context.

JSON logs are also saved locally in backend/logs/sessions/ for quick debugging without cloud access.

Update your .env file:

GOOGLE_API_KEY=your_api_key
ENABLE_CLOUD_TRACE=true
GOOGLE_APPLICATION_CREDENTIALS=certs/albert-logger-GCP-key.json

• Python 3.9+
• Node.js & npm
• Google Account (for Gmail)

1. Clone the repository:

   git clone https://github.com/nomadmanhattan/Albert_Public_Demo.git
   cd Albert_Public_Demo

2. Backend Setup:

   cd backend
   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   pip install -r requirements.txt
   playwright install chromium

3. Frontend Setup:

   cd ../frontend
   npm install

4. Environment Variables: Create a .env file in backend/ with your Google Cloud credentials (if using GCS) or Gemini API key.

   GOOGLE_API_KEY=your_api_key_here

Important: For the first run, you must run the backend locally (see below) to sign in to Google. This creates a session file that Docker will use.

1. Run Locally Once: Follow "Running Locally" steps to sign in.
2. Start with Docker:

   docker-compose up --build

3. Open your browser to http://localhost:3000.

1. Start the Backend:

   # In backend/ directory
   source venv/bin/activate
   uvicorn main:app --reload

2. Start the Frontend:

   # In frontend/ directory
   npm run dev

3. Open your browser to http://localhost:5173.

1. Say Hello: Tell Albert what you want, e.g., "Make me a podcast about AI updates in the last 3 days".
2. First Run: Albert will ask you to log in to Gmail if you haven't already.
3. Listen: Wait a moment, and Albert will give you a link to your generated audio overview!

MIT