IIIT Hyderabad’s Subtl.ai V2 Outperforms OpenAI’s RAG Solutions

Subtl.ai, a startup born out of the Indian Institute of Information Technology Hyderabad, has unveiled ‘Subtl V2’, the latest iteration of its AI information retrieval system for enterprises. 

Vishnu Ramesh, founder of Subtl.ai, took to LinkedIn to announce the launch. He said that their auto-tuning pipeline built by engineers and researchers at IIIT Hyderabad outperforms RAG built on OpenAI and open source embeddings by 15-20%. 

Ramesh also announced that their ‘co-create embeddings program’ is now available, which allows developers to create proprietary RAG embedding models based on private documents. 

Furthermore, Subtl.ai can now also generate an Excel-like grid view for data and said that document size and page limits have been removed from the app. 

Founded in 2020, Subtl.ai has built a local solution to derive insights and analytics from private and sensitive data in enterprise applications. In a blog post earlier this year, the startup revealed that one of India’s biggest banks, State Bank of India, successfully implemented Subtl.ai. The results demonstrated 92% accuracy in information retrieval, and 56,570 minutes were saved, which was equivalent to 5 lakh rupees. 

In an interview with AIM earlier this year, Ramesh revealed that the startup’s ambition was to reduce the dependence on companies like OpenAI. 

“Microsoft has a lot of security features on top of its OpenAI offerings. But at the end of the day, it’s still a common endpoint, making it difficult for companies in the banking or sensitive data sector to rely on it,” he said. 

Ramesh also called Subt.ai a ‘private Perplexity’ built on light models for the enterprise. The tool is built on the Llama 3 8B variant and can access all the information in catalogues and databases that private companies have trained on. Moreover, it can also reply to simple questions citing the exact source along with a paragraph.

Subtl.ai isn’t free from competition, and companies like Google, Anthropic, and OpenAI offer retrieval capabilities through its API. On the other hand, Perplexity also released Internal Knowledge Search, a pro version for enterprises. However, enterprises are sceptical about their private data going online, especially given the Amazon Q fiasco.