Within a day of OpenAI’s announcement that GPT-3.5 was available for fine-tuning with custom data, artificial intelligence consumer finance pioneer Prodigal was already using it to enhance their proprietary AI Intent Engine.
“OpenAI released the ability to fine-tune GPT-3.5 on a Thursday and we were able to create fine-tuned models for it by Friday. We are perhaps one out of a thousand companies around the world innovating at that speed and using it to improve our products,” said Akshat Vaidya, Product Manager at Prodigal.
Being able to implement fine-tuned GPT-3.5 within Prodigal’s custom-trained AI solutions immediately delivered improvements for their customers.
Having been in the AI space for years, Prodigal’s AI team has already made use of numerous large language models, including XGBoost classifiers, BERT-based models, and LLaMA 2.
And now adding OpenAI’s GPT-3.5 to the mix has only strengthened Prodigal’s models.
“With fine-tuning we can bring all of the millions of calls that we have, that every finance organization has, to bear. We can learn from the best in every organization and we can bring that together to build really high quality models,” explained Harsh Pareek, Senior ML Engineer at Prodigal.
“It’s the marriage of unique proprietary data and unique proprietary technologies,” said Shantanu Gangal, Prodigal’s CEO.
Training LLMs with Prodigal’s extensive custom data has allowed the team to build a suite of solutions specifically for consumer finance. Coupling data curated from over 300 million lending and borrowing interactions with the power of LLMs delivers the ability for Prodigal’s customers to extract new insights from the conversations they have with customers every day.
When GPT-3.5’s fine-tuning abilities were announced, the team was initially apprehensive. While OpenAI “has the best models in the industry,” Pareek said, not being able to train the model on custom data was a real drawback. “Now, with fine-tuning on Prodigal’s proprietary data, you can get performance exceeding that of GPT-4 at really fast speeds.”
The possibilities for consumer finance with Prodigal’s fine-tuned models benefit both customers and companies.
“The borrowing and lending industry has been following these workflows which are ripe to be supercharged by generative AI. We’re still very stuck in our archaic world where everything is template-based. With generative AI and the advent of LLMs, consumer finance teams are going to be able to deliver personalization to customers at a scale that was previously unheard of,” Vaidya said.
“Our primary goal is to build applications that give creditors actionable insights into consumer finances. Change is the only constant. Technology is evolving rapidly and new paradigms are opening new horizons. Fine-tuned GPT-3.5 is just one illustration of it. It allows blending proprietary data to improve insights. We always bring these improvements to our customers quickly,” concluded Prodigal CEO Gangal.