Generative AI system open-source model customization that resulted in a reduction in operational costs for a financial technology company.

Challenge

A fintech company needed to automate customized call transcript analysis and deliver personalized recommendations to its clients. Initially, they were using a high-cost GPT-4-based model, which significantly increased operational expenses. They sought a more cost-effective solution without compromising on performance. 

Solution

We developed a custom generative AI system by leveraging prompt engineering and agent workflows to create a knowledge base for document analysis. Additionally, we implemented instruction fine-tuning on an open-source LLM, replacing GPT-4 with a fine-tuned model tailored to the client’s specific needs. This allowed the client to maintain high-quality content generation while reducing costs. The new system facilitated the production of personalized marketing materials based on customer preferences. 

Result

The solution not only saved the client 20 hours per week in manual call reviews but also reduced operational costs by 90% through the switch from GPT-4 to an open-source model. The client achieved higher engagement rates while significantly lowering the ongoing expenses of running their AI system.