Smart Mail Implementation for the Financial Sector

Solutions

• The raw data from the client contained inconsistencies, missing values, or irrelevant information. Hence, preprocessing steps are carried out to ensure the quality of the dataset.
• The preprocessing steps include setting up an Amazon SageMaker instance, understanding the data, and writing a script for data acquisition, data cleaning, data transformation, and validation.
• Once data preprocessing is completed, we store the processed data in Amazon DynamoDB, which serves as the customer data for the model.
• Building the automation to fetch the latest data from the customer database using the AWS lambda function.
• Using Power Automate, which enables us to create automated flows, two flows were created to retrieve newly arrived emails and reply to them.
• Amazon API Gateway is integrated with AWS Lambda to trigger the new mail arrival flow in Power Automate through an API.
• A validation layer was added in AWS Lambda to validate the customer email and verify the customer using PAN Number or FOLIO number using the customer data stored in Amazon DynamoDB.
• For text generation to respond to customer emails, we have chosen the “Anthropic Claude v2” model to generate human-like text in response to the customer’s email.
• Parameters are set to optimize the model’s performance and reduce randomization and probability for text generation.
• We have created an Amazon DynamoDB Table to record the previous conversation using the conversation ID of the email chain so the model can find the point of continuity and respond to the customer’s continuous queries using the previous conversation.
• Configured AWS Lambda function to make API calls to the model to generate the response of the email and process prior to transmitting responses to the reply flow in Power Automate through REST API.