6 months ago
44
While generative AI may not appear on some enterprises’ balance sheets yet, that doesn’t mean it can’t generate revenue. Shorthills AI, a startup based in Gurgaon that helps enterprises build fine-tuned models, is on track to generate $5 million this year.
Founded in 2018 by Paramdeep Singh and Pawan Prabhat– the startup– which refers to itself as a generative AI transformation company– is helping enterprises make sense of unstructured data.
Unstructured data remains one of the major challenges for enterprises. However, if enterprises can effectively analyse and interpret this data, they can derive a lot of value.
One of the startup’s strengths lies in assisting enterprises in developing search capabilities powered by generative AI, enabling them to sift through millions of documents and extract valuable insights.
( Pawan Prabhat & Paramdeep Singh, co-founders at Shorthills AI)
Making Sense of Unstructured Data
Paramdeep Singh, president and co-founder at Shorthills AI, told AIM that many enterprises need help with financial data, which often is unstructured, such as emails, documents, and scattered databases.
“In an enterprise, financial information can be scattered across various platforms—some in SAP, some in emails, and others in Salesforce. To address this, we have developed engines capable of navigating these different data sources to extract insights and create value,” he revealed.
One of the startup’s key offerings is building search capabilities that can scan through these data points and provide insights and actionable recommendations, helping businesses make informed decisions based on their financial data.
The startup has also built capabilities around legal documents for some customers. “We have built models to analyse legal contracts to extract key details such as the signing date, amounts owed by different parties, and specific clauses. By gathering this information from the documents, we enable businesses to leverage these insights as key differentiators,” Singh said.
AI Model for Finance
The co-founders also revealed that the startup is building a Language Model specifically for the financial domain. “We are trying to build our own from scratch on bare metal. We have purchased our own Graphics Processing Units (GPUs),” Pawan Prabhat, co-founder at Shorthills AI, told AIM.
When asked if the model is similar to a large language model (LLM) like Llama 3, Prabhat revealed that while the architecture remains the same, Shorthills AI’s model is more domain-specific. Hence, it will be smaller as well.
While the startup has yet to release its model, it is already helping enterprises make sense of unstructured data. For some of its customers, the startup has built applications on top of open-source models such as Llama 3. Nonetheless, Prabhat also revealed that its approach is model agnostic, which means it can leverage open-source models like Llama or closed-source models like OpenAI’s GPT-4, depending on the use case.
Earlier this year, PwC announced its plans to integrate Shorthills AI’s generative AI features into its suite of tools. Through this partnership, PwC will implement Shorthills AI’s Generative AI solutions, which include AI-powered search, automatic classification of entries, document summarisation, and natural language-based Q&A.
Data and AI Service
The AI landscape is rapidly evolving. What is working today might become obsolete tomorrow. Such is the pace of development. When AIM asked the co-founders what would happen if enterprises opted to develop these AI capabilities in-house instead of relying on third parties, they noted that while some organisations might have the resources and technical expertise, many still need to.
“A typical manufacturing company’s primary strength lies in manufacturing, not in developing a large language model to analyse their legal databases. However, they must manage legal databases and respond to all RTIs and notices. Therefore, they need a solution effectively addressing these challenges,” Singh said.
Additionally, Prabhat stressed that fine-tuning a model involves a solid understanding of data engineering and how data pipelines operate. Moreover, the challenges don’t end with implementing a single model.
“Each new model changes calibration and structure, so we must build a flexible framework to adapt to these evolving models. If a new model emerges that offers superior accuracy, we must be prepared to integrate it seamlessly,” he pointed out.
Competing with Indian IT?
Interestingly, many of the IT services company’s Proof-of-Concept (PoC) involve building fine-tuned models for enterprises trained on their internal data. This puts the startup directly in partnership with these IT services companies.
“I would assume that they would get into something similar. So Accenture seems to be doing something around this. And I’m sure most IT services companies would eventually start doing something around this,” Singh said.
However, according to him, the critical difference is in the focus. Most IT services companies have not disclosed whether they are generating revenue from generative AI. Even for those who have, many industry experts remain sceptical.
A significant chunk of these IT services companies’ business still involves doing IT services. “I believe they face a larger battle at this time. For a company like ours, which is intensely focused on AI and data, we’re concentrating on that current 1% of spending. While I’m not suggesting they will abandon other areas, they shouldn’t be IT firms’ primary focus right now,” Singh concluded.
