In May, the Department of Science & Technology (DST) announced the launch of a new hub dedicated to creating Indic language models. This new hub, BharatGPT, was created in collaboration with IIT Bombay, IIT Madras, IIT Hyderabad, IIIT Hyderabad, IIM Indore, and IIT Mandi.
The initiative aims to develop LLMs in Indian languages for India, along with applications for Indian enterprises.
Apart from working on BharatGPT, Ganesh Ramakrishnan, a professor at IIT Bombay, has been dedicated to developing translation engines. To continue this bid, he has co-founded bbsAI with Ganesh Arnaal, which has been a decade in the making.
“Arnaal approached me in 2013 with the idea of developing a translation engine to translate technical books from English into Hindi and other major Indian languages. Thus, the Udaan Translation Project was born,” Ramakrishnan recalled in an exclusive interaction with AIM.
At the recent Global INDIAai Summit 2024, Ramakrishnan discussed how AI can produce groundbreaking outcomes for real business applications in data-scarce environments. He underscored the significance of creating small language models and innovating algorithms.
Emphasising on human centricity and inclusive AI, Ramakrishnan added that the approach of making small language models for Indic languages addresses the challenge of limited data, enabling the delivery of dependable and practical solutions to the industry. This has led to the founding of bbsAI.
Initially funded by Arnaal, bbsAI officially became a commercial entity in February 2023, entering a licence agreement with IIT Bombay for the commercial exploitation of the Udaan Translation Engine.
Flying with Udaan
The journey for bbsAI started with the Udaan, which stands out in the crowded market of translation tools. Ramakrishnan explained, “Our engine is a result of training models that are probabilistic in nature, but we introduced technical dictionaries as constraints to overcome hallucinations and inaccuracies.”
This deterministic approach, powered by their own open-sourced data-efficient machine learning algorithms and grounded in extensive language resource research by Arnaal, ensures accurate and context-appropriate translations in scientific and technical fields. “The Udaan Translation Engine offers a comprehensive ecosystem: an OCR engine preserving the source document’s style and layout, a translation engine, and a user-friendly post-editing tool.”
In 2022, Ramakrishnan and Arnaal met education minister Dharmendra Pradhan, who appreciated their dedication to building Udaan. “They have developed a translation tool— Udaan—that is breaking the language barrier in education by translating learning materials in Indian languages,” the minister tweeted.
Revolutionising the Insurance Industry
Expanding its offerings to leverage its digitalisation and OCR capabilities, bbsAI has introduced a suite of AI-enhanced process automation solutions that has the potential for a variety of use cases across industries.
“As a natural extension of the machine learning capabilities we have built over the years, we have begun to offer process automation solutions by building small language models that can provide intelligent, accurate and inherently deterministic solutions to automate a variety of business processes,” Ramakrishnan elaborated.
bbsAI developed an AI solution for ICICI Lombard’s quotation management system (QMS). “Our solution captures data from various file formats and populates it automatically into the templated underwriting formats, delivering productivity gains,” said Ramakrishnan.
This solution is a global first in the insurance industry, achieving over 90% accuracy while adhering to strict data privacy regulations with limited datasets. “We have delivered a staggering accuracy of over 90% while completely eliminating hallucinations,” he emphasised.
Small Language Models and Explainability
Ramakrishnan explained bbsAI’s unique approach, which is built on small language models and explainability by design.
“LLMs perform many tasks, but for business use-cases, explainability and reliability are crucial,” Ramakrishnan stressed. This focus on deterministic solutions has enabled bbsAI to create accurate, reliable, and explainable AI solutions, fostering greater industry adoption.
“We integrate domain knowledge and cross-industry understanding as an integral part of the development process, not as an afterthought,” he added.
Moving Beyond POCs
One of bbsAI’s significant milestones is its transition from proof of concept (PoC) to real-world AI solutions. “The key is shifting from probabilistic to deterministic models, providing explainable and accurate solutions,” noted Ramakrishnan.
This approach has not only inspired user confidence but has also demonstrated tangible benefits in efficiency and productivity for clients. “With our unique approach, we have successfully converted AI promises into products and solutions,” he asserted.
bbsAI’s journey from a visionary project to a trailblazer in business automation and translation technology is truly remarkable. “We at bbsAI are passionate about making technology available to all Indians,” added Ramakrishnan.
Bharat Bhasha Sanganan
At its core, bbsAI is driven by the vision of Bharat Bhasha Sanganan, meaning Indian language computing. “In India, only those who know English have privileged access to technology. If we look globally, most developed nations have access to technology in their native languages,” Ramakrishnan explained.
bbsAI (which stands for Bharat Bhasha Sanganan AI) has taken significant steps to bridge this gap, starting by creating a complete Hindi user interface for LibreOffice, bbsहिन्दीoffice and is planning to extend this to other major Indian languages.
bbsAI has a natural synergy with the National Education Policy (NEP), which has catalysed higher learning through Indian languages, aligning perfectly with bbsAI’s mission.
“From the academic year 2023-24, engineering and medicine are being taught in 11 Indian languages,” Ramakrishnan mentioned. This shift is expected to boost the demand for textbooks in Indian languages, making bbsAI a valuable partner for publishers and academic institutions.
“We have been working on machine translation for technical domains for over a decade, ensuring the use of domain-specific vocabulary in our translations,” he concluded.
