OnFinance, a Bengaluru-based finance startup is leveraging AI to bring solutions for the banking, financial services and insurance (BFSI) sector. NeoGPT, the company’s proprietary LLM is specifically built to cater to analysts, advisors, and BFSI companies.
“Recently, we became the first GenAI startup to have NSE [National Stock Exchange of India] as a client,” said Anuj Srivastava, co-founder and CEO of OnFinance, in an exclusive interaction with AIM.
ChatGPT for Enterprise Finance
OnFinance’s use cases cover an entire gamut of financial services for enterprises in the form of financial research, equity research, AI copilots for underwriting, and index research. Furthermore, there’s relationship management, compliance and AI copilots for wealth advisory. “Think of it as ChatGPT, but for enterprise finance,” said Srivastava.
Working within the critical finance sector, where customers are wary of their data being used for AI, OnFinance ensures that the criticality is maintained. Srivastava points this out as the primary reason for enterprises to shy away from ChatGPT APIs.
“All the financial clients in India currently want on-prem solutions because they don’t want their data to go out of their cloud or infra. That’s why they are still not able to use OpenAI, APIs and so on,” said Srivastava.
To ensure data security, the company ensures that no customer data is used for training. “We don’t use any data for training, fine tuning or even feedback loop analysis. We actually pass the data on a real time basis to a copilot, so that, you know, it’s not used for those purposes.”
Accelerating AI Startups
With the option of GPT Stores and similar models, where companies can train and fine-tune their own models with their proprietary data, finance startups such as OnFinance find better relevance.
“What happens is that they [models built on GPT Store] are not scalable, and can’t be productionised. The main reason why they are not implementable is that you can’t directly access those APIs after building a GPT. Even if you are able to access them, they have a rate limit, so you won’t be able to productionise them,” said Srivastava.
“In GPT Stores, any data or question that your analyst would ask, will go directly to OpenAI,” quipped Srivastava.
Furthermore, big players such as Microsoft will never be a direct threat to these companies simply because it only boosts their business. Srivastava explains how, just like them, there would be 20-40 companies building domain-specific models and would approach Azure for cloud and GPU compute.
“So the revenue that they would be able to earn from these 40 companies would be way higher than from building their domain-specific model and selling them as copilots,” he said.
Interestingly, the startup has cloud partnerships with AWS, Google and Microsoft Azure. NeoGPT is also available on AWS Bedrock.
Open Source is Still The Way
While data security is a priority, OnFinance’s proprietary models are built on open source models that are publicly available, namely Code Llama by Meta AI and Mistral 7B by Mistral AI. “On top of these two models, we fine-tune a lot of financial data sets such as AGM reports, credit reports, compliance circulars, and more,” said Srivastava.
In addition, instruction tuning is done to the Q&As formulated with the information of 25,000 news websites related to finance, including Moneycontrol, Financial Times, and others. These datasets are made on a Q&A basis. “So, that’s why it’s very intelligent when it comes to financial data,” he said.
Partnerships and JioGenNext
OnFinance has already partnered with 15 of the biggest banks and wealth management players in India, including ICICI and exchange platform Centrum. It recently tied up with NSE, Oister Global, and Let’s Venture.
The startup is also part of JioGenNext, the renowned startup accelerator program that promotes GenAI startups. Through this program, OnFinance will be able to provide their services to big tech clients of JioGenNext.
Srivastava, a BITS Pilani alum, co-founded OnFinance with Priyesh Srivastava, who is also from the same college. Last November, the startup received a seed funding of $1.1 million, and both founders were named in the Forbes 30 Under 30 Asia list.
