We empowered a leading financial institution with an AI-driven Conversational Platform on AWS, reducing product information retrieval time by 82% and enhancing customer engagement by 67%

CHALLENGE

The financial institution's RMs face significant hurdles in managing product information and responding to client queries efficiently, encountering the following challenges:

• RMs spend excessive time navigating through extensive product documentation to locate specific information, leading to delayed response times and reduced productivity
• The absence of a centralized, intelligent knowledge retrieval system makes it challenging to access accurate product details quickly and confidently
• Manual search processes increase the risk of overlooking crucial information or providing incomplete responses to client queries
• With a vast portfolio of mutual fund products and complex SEBI guidelines, ensuring consistent accuracy and compliance in responses becomes increasingly challenging

Why were we brought in?

Ganit has made a significant dent in various industries using data science and analysis. Ganit partners with clients to translate their data into a tangible, insightful plan of action that delivers on a measurable impact to the clients’ topline & bottom-line growth.

With a strong Gen AI practice and a dedicated Gen AI Centre of Excellence (CoE), Ganit has advanced capabilities in building consumable Gen AI driven solutions for our customers. With almost 20 demos catering to different industry verticals and business functions, Ganit has a proven track of delivering customer delight through accelerated implementation.

Solution

Ganit developed a conversational AI platform on AWS that transforms how RMs access and utilize product information. The solution combines intelligent document processing with real-time query handling for accurate, SEBI-compliant responses.

• The solution is built around a knowledge base of 5,000+ HDFC AMC mutual fund documents stored in S3, serving as the core data for the RAG chatbot. The infrastructure uses four distinct S3 buckets for knowledge base, chat history, React application, and ECS packages
• For user interaction, a React application is hosted on S3 and served through CloudFront CDN. Users access the system through Cognito authentication, where they can submit either text or voice queries (processed via Transcribe). The system is dimensioned for 20 daily users
• Document processing is managed through ECS on Fargate, which runs containerized tasks to process new S3 uploads. These documents are chunked and transformed into vectors using Cohere's multilingual model, then stored in PostgreSQL with the pgvector extension
• The query processing flow begins when users submit queries through API Gateway, triggering Lambda functions. These functions, supported by Lambda layers from Docker images, execute the RAG process - retrieving relevant context from PostgreSQL, combining it with the query, and generating responses through selected Claude models
• Responses are delivered within 1-3 seconds and include citations for transparency. All user interactions and chat history are stored in DynamoDB and S3, with optimized storage and caching systems configured for efficient performance

Features of the tool

• The AI-powered conversational platform seamlessly integrates comprehensive product knowledge with natural language processing capabilities, enabling instant and accurate information retrieval
• The solution ensures SEBI compliance while providing user-friendly, conversation-style interactions, making complex product information easily accessible to RMs