Build a Banking Workflow App Using Django, MongoDB & Voyage AI (Datacamp)

A hands-on codealong project that demonstrates how to build a banking document workflow application with semantic search and retrieval augmented generation (RAG).

Key Takeaways

• Build a functional web application with Django to serve as your workflow interface.
• Implement semantic search and retrieval augmented generation using MongoDB Atlas Vector Search and Voyage AI.
• Get hands-on with a practical banking workflow use case, from document retrieval to automated task handling.

Prerequisites

• Python 3.12+
• A MongoDB Atlas cluster (free tier works)
• A Voyage AI API key

Setup

1. Clone the repository and checkout the starter branch:

   git clone <repo-url>
   cd banking-workflow-codealong
   git checkout starter

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Create a .env file in the project root:

   SECRET_KEY=your-secret-key-here
   DB_ENGINE=django_mongodb_backend
   MONGODB_URI=mongodb+srv://<username>:<password>@<cluster>.mongodb.net/?appName=bankflow
   DB_NAME=banking_workflow
   VOYAGE_API_KEY=your-voyage-ai-key
   

5. Run migrations:

   python manage.py migrate

6. Start the development server:

   python manage.py runserver

7. Visit http://127.0.0.1:8000/ in your browser.

Project Structure

banking/                  # Main application
├── models.py            # BankingDocument & WorkflowTask models
├── views.py             # View functions for all pages
├── forms.py             # Django forms
├── urls.py              # URL routing
├── admin.py             # Admin configuration
├── services/
│   ├── embedding_service.py      # Voyage AI embeddings
│   ├── vector_search_service.py  # MongoDB Atlas Vector Search
│   └── rag_service.py            # RAG pipeline
└── templates/banking/   # HTML templates

Branches

• main — Complete solution with all features implemented
• starter — Skeleton with TODOs for the codealong

Atlas Vector Search Index

Create this index on the banking_bankingdocument collection in your Atlas cluster.

Index name: banking_embedding_index

{
  "fields": [
    {
      "type": "vector",
      "path": "embedding",
      "numDimensions": 1024,
      "similarity": "cosine"
    }
  ]
}

UI Screenshot: