Personalized Banking Services

A Python-based system that provides personalized banking product recommendations based on user data, spending patterns, and social media activity.

Project Structure

personalised_banking_services/
├── data/                      # Data directory
│   ├── Account_Statement.csv              # User's bank transactions
│   ├── credit_card_transactions.csv       # Credit card transaction history
│   ├── KYC_Details.csv                    # Know Your Customer details
│   ├── social_media_posts.csv             # User's social media activity
│   ├── emails_to_banks.csv          # Email communications
│   ├── Receiver_vs_Category.csv           # Transaction categorization
│   ├── Credit_Card_Details.csv# Available credit card products
│   ├── Loan_Details.csv       # Available loan products
│   └── credit_card_list.csv               # User's existing credit cards
├── src/
│   ├── ai/
│   │   ├── llm_interaction.py    # LLM-based recommendation generation
│   │   └── llm_analyzer.py       # LLM-based analysis of user data
│   ├── analysis/
│   │   └── financial_analyzer.py # Financial data analysis
│   ├── data_processing/
│   │   ├── data_loader.py        # Data loading and preprocessing
│   │   └── data_extractor.py     # Feature extraction and analysis
│   └── config.py                 # Configuration settings
├── output/                    # Generated recommendations and analysis
│   ├── spending_analysis.json
│   ├── kyc_details.json
│   ├── user_interests.json
│   ├── product_recommendations.json
│   └── credit_card_recommendations.json
├── main.py                    # Main script
└── README.md                  # This file

Features

1. Data Processing

   • Transaction analysis
   • Spending pattern recognition
   • Social media interest extraction
   • KYC information processing

2. Recommendation Generation

   • Credit card recommendations
   • Loan recommendations
   • Other financial product suggestions
   • Personalized based on user profile

3. Real-time Updates

   • Update recommendations with new social media posts
   • Maintain historical data
   • Incremental updates

🛠️ How We Built It

1. Data Processing Layer:

   • Financial data loading and processing
   • KYC details extraction
   • Spending pattern analysis
   • User interest identification

2. Python 3.8+

3. Required packages:

   pandas>=2.0.0
   numpy>=1.24.0
   scikit-learn>=1.3.0
   llama-cpp-python>=0.2.0
   sentence-transformers>=2.2.0
   

LLM Model Details

This project was developed and demoed using:

• Mistral 7B Instruct v0.2 (Q4_K_M quantized version)
• Download from Hugging Face: mistral-7b-instruct-v0.2.Q4_K_M.gguf
• Place in the models directory

Setup Instructions

1. Clone the Repository

   git clone https://github.com/yourusername/personalised_banking_services.git
   cd personalised_banking_services

2. Create and Activate Virtual Environment

   # Create virtual environment
   python -m venv venv
   
   # Activate virtual environment
   # On macOS/Linux:
   source venv/bin/activate
   # On Windows:
   .\venv\Scripts\activate

3. Install Dependencies

   pip install -r requirements.txt

4. Download Required Models

   # Create models directory
   mkdir -p models
   
   # Download sentence transformer model
   python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('all-MiniLM-L6-v2').save('models/sentence_transformer')"
   
   # Download Llama model (example using 7B model)
   # Note: You'll need to download the model file manually from Hugging Face
   # and place it in the models directory

5. Prepare Data Directory

   # Create data directory if it doesn't exist
   mkdir -p data

6. Configure Data Files Place the following CSV files in the data directory:

   • Account_Statement.csv - Bank transaction history
   • credit_card_transactions.csv - Credit card usage
   • KYC_Details.csv - Personal information
   • social_media_posts.csv - Social media activity
   • emails_to_banks.csv - Email communications
   • Receiver_vs_Category.csv - Transaction categorization
   • Credit_Card_Details.csv - Available credit cards
   • Loan_Details.csv - Available loan products
   • credit_card_list.csv - Existing credit cards

7. Configure Settings Update the src/config.py file with your settings:

   # Base directory paths
   BASE_DIR = Path(__file__).parent.parent
   DATA_DIR = BASE_DIR / "data"
   OUTPUT_DIR = BASE_DIR / "output"
   
   # Data file paths
   DATA_FILES = {
       "transactions": DATA_DIR / "Account_Statement.csv",
       "credit_card_transactions": DATA_DIR / "credit_card_transactions.csv",
       "social_media": DATA_DIR / "social_media_posts.csv",
       "kyc": DATA_DIR / "KYC_Details.csv",
       "emails": DATA_DIR / "emails_to_banks.csv",
       "receiver_categories": DATA_DIR / "Receiver_vs_Category.csv",
       "credit_cards": DATA_DIR / "Credit_Card_Details.csv",
       "loans": DATA_DIR / "Loan_Details.csv",
       "credit_card_list": DATA_DIR / "credit_card_list.csv"
   }
   
   # Create output directory if it doesn't exist
   OUTPUT_DIR.mkdir(exist_ok=True)

8. Verify Setup

   # Run a test analysis
   python main.py --test

Data Format Requirements

1. Account Statement CSV

   Date,Description,Amount,Category
   2024-03-01,Grocery Store,-50.00,Food
   2024-03-02,Salary,3000.00,Income

2. Credit Card Transactions CSV

   Date,Merchant,Amount,Category
   2024-03-01,Amazon,-100.00,Shopping
   2024-03-02,Gas Station,-40.00,Transportation

3. KYC Details CSV

   Field,Value
   Name,John Doe
   Age,30
   Income,75000
   Employment,Full-time

4. Social Media Posts CSV

   Date,Platform,Content,Engagement
   2024-03-20,Twitter,Post content here,15
   2024-03-20,Instagram,Another post,25

Troubleshooting

1. Model Download Issues

   • Ensure sufficient disk space (at least 10GB)
   • Check internet connection
   • Verify write permissions in the models directory

2. Data Loading Errors

   • Verify CSV file formats
   • Check file permissions
   • Ensure all required files are present

3. Memory Issues

   • Reduce batch size in config.py
   • Use smaller model variants
   • Close other memory-intensive applications

Running the Scripts

1. Full Pipeline

To run the complete analysis pipeline:

python main.py

This will:

• Load all data from the data directory
• Process transactions and user information
• Generate recommendations
• Save results to the output directory

2. Update with New Social Media Posts

To update recommendations based on new social media activity:

python main.py --update-social path/to/new_posts.csv

The new posts CSV file should have the following format:

Date,Platform,Content,Engagement
2024-03-20,Twitter,Post content here,15
2024-03-20,Instagram,Another post,25

System Workflow

1. Data Collection and Loading

   • System loads various data sources from the data directory
   • Transaction data from bank statements and credit cards
   • KYC information and user details
   • Social media posts and engagement metrics
   • Available banking products information

2. Data Processing Pipeline

   • data_loader.py:
     • Handles loading and initial preprocessing of all data sources
   • data_extractor.py:
     • Extracts relevant features and patterns from the data
     • Validates data format and completeness
     • Handles missing values and data cleaning

3. Analysis Phase

   • financial_analyzer.py:

     • Analyzes spending patterns and transaction history
     • Identifies spending categories and trends
     • Calculates financial metrics and ratios
     • Generates spending analysis reports

   • llm_analyzer.py:

     • Processes social media posts and user communications
     • Extracts user interests and preferences
     • Analyzes sentiment and engagement patterns
     • Identifies lifestyle indicators

4. Recommendation Generation

   • llm_interaction.py:
     • Combines insights from financial and social analysis
     • Matches user profile with available products
     • Generates personalized recommendations
     • Provides reasoning for each recommendation

5. Output Generation

   • Saves analysis results in JSON format
   • Creates detailed recommendation reports
   • Maintains historical data for trend analysis

6. Update Process

   • When new social media posts are added:
     • Only processes the new data
     • Updates user interests and preferences
     • Regenerates recommendations without full pipeline rerun
     • Maintains consistency with existing analysis

Output Files

1. spending_analysis.json

   • Total spending
   • Category-wise spending
   • Top merchants
   • Monthly trends

2. kyc_details.json

   • User profile
   • Income details
   • Updated interests and hobbies

3. user_interests.json

   • Extracted interests from social media
   • Transaction-based interests

4. product_recommendations.json

   • Recommended credit cards
   • Loan suggestions
   • Other financial products
   • Reasons for recommendations

5. credit_card_recommendations.json

   • Detailed credit card recommendations
   • Card benefits
   • Annual fees
   • Interest rates

Error Handling

The system includes comprehensive error handling for:

• Missing files
• Invalid data formats
• Processing errors
• Data validation failures

Error messages are color-coded for better visibility:

• 🔵 Blue: Processing information
• 🟢 Green: Success messages
• 🟡 Yellow: Warnings
• 🔴 Red: Errors