Modular LLM Workflow Engine

A language processing system with modular workflows and LLM-agnostic architecture for product information handling.

🚀 Installation

Linux/MacOS Setup

# Clone repository
git clone https://github.com/your-repo/product-workflows.git
cd product-workflows

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -e .

# Install langgraph development tools (optional)
pip install "langgraph[dev]"

⚙️ Configuration

Create .env file in project root:

PROJECT_ID="your-gcp-project"
LOCATION="your-gcp-region"
STAGING_BUCKET="gs://your-bucket"
LLM_PROVIDER="vertexai"  # or "openai"

📖 Basic Usage

from core.workflows.app import SimpleLangGraphApp

# Initialize with VertexAI
app = SimpleLangGraphApp(
    project="your-project",
    location="us-central1",
    provider="vertexai"
)
app.set_up()

# Execute product query
response = app.query("Get details for premium headphones")
print(response)

🏗 Project Structure

src/
├── core/
│   ├── nodes/           # Processing components
│   │   ├── routing.py    # Message routing logic
│   │   ├── tools/        # Custom tools
│   │   └── output.py     # Response formatting
│   └── workflows/       # Workflow definitions
│       └── product_workflow.py  # Main workflow graph
├── providers/           # LLM integrations
│   └── llm_factory.py   # Provider switching
├── config/              # Environment setup
├── main.py              # Main entry point for CLI execution
└── agent.py             # Entry point for langgraph dev server

🔄 Workflow Flow

1. Input Reception → 2. Model Routing
       ↓
4. Output Formatting ← 3. Tool Execution

Key processing stages:

1. Model Routing: LLM decides between direct response or tool usage
2. Tool Execution: Product details retrieval from knowledge base
3. Response Formatting: Final output standardization

✨ Features

• LLM Flexibility: Switch between VertexAI/OpenAI via config
• Modular Design: Add new nodes without affecting existing flows
• Structured Outputs: Consistent formatting for API consumption
• Tool System: Extendable product information base
• Local Testing: Console interface for workflow validation

🛠 Development

# Run main application (CLI mode)
python -m src.main

# Run interactive development server
langgraph dev src.agent:graph

# Execute tests
pytest tests/ -v

# Generate coverage report
coverage run -m pytest tests/
coverage report

Entry Points

The project has two main entry points:

1. src/main.py: Command-line interface for running predefined test queries

   • Use this for quick testing and batch processing
   • Run with python -m src.main

2. src/agent.py: Exposes the workflow graph for the LangGraph development server

   • Use this for interactive debugging and visualization
   • Run with langgraph dev src.agent:graph
   • Access the web interface at http://localhost:3000

📜 License

Apache 2.0 - See LICENSE for details