×
Case Studies Blog Careers Technologies
Assisted Sales: Harnessing Generative AI to Enhance Interactions Between Field-level Sales Representatives and Prospective Customers
Industry
Banking
Technologies
Python, AWS Sagemaker, Meta Llama, Amazon Textract, Mixtral
About Client

Our client is a prominent member of one of India's top 5 Financial Services Conglomerates, holding a leading position in the Home Loan industry within the country.

Problem Statement

The client faced a significant challenge in the lengthy and time-consuming process of educating and validating new leads. This involved extensive efforts from all field agents who had to navigate through extensive Credit Guidelines for each Home Loan policy. These guidelines, sometimes spanning hundreds of pages, posed a considerable obstacle in responding to lead queries and validating them efficiently, resulting in a process that took several weeks.

Solution

To address this challenge, our team, in collaboration with the client, initiated a Proof of Concept (PoC) stage. Initially, the client provided five credit guideline documents, each corresponding to a different policy. These documents contained interactive content, prompting our team to enhance the RAG architecture. Leveraging advanced language models like Llama2-70B and Mixtral-8x7B, we developed a solution that could generate precise answers to various questions posed by the client’s policy team.

Upon the successful completion of the initial PoC stage, the client extended the collaboration by providing another set of five documents. These documents were more intricate, featuring complex content such as tables with multiple merged cells that spanned several pages with or without headers.

The Oneture team rose to the challenge, further enhancing the RAG architecture to comprehend and extract this intricate content. This phase proved to be crucial and intricate, as existing Python libraries and AWS services like Textract struggled to extract data from the complex table structures. Our solution involved a custom code that initially extracted the structure using Textract and then utilized custom Python code to organize the data into the updated structure specified in the documents.

Technologies
  • Development: Python
  • Language Models: Llama2-70B, Mixtral-8x7B
  • AWS Services: Sagemaker Notebooks, Textract
Value Delivered

The successful completion of the PoC marked a significant milestone in our collaboration. Currently, we are actively engaged with the client in defining the next steps of the project. This solution not only streamlined the lead validation process but also showcased the effectiveness of advanced Gen AI technologies in overcoming intricate challenges related to lengthy sales process