Historic leap: IIT Roorkee’s AI MoScNet decodes Modi script for heritage preservation

IIT Roorkee has developed the world’s first artificial intelligence model capable of transliterating the ancient Modi script into Devanagari, offering a major breakthrough in the digital preservation of India's historical manuscripts.

The AI tool, named MoScNet, converts handwritten texts into readable Devanagari script, making centuries-old documents accessible for researchers, educators, and archivists.

Part of a larger initiative called Historic Scripts to Modern Vision, the project also includes a unique dataset named MoDeTrans.

The collection features more than 2,000 verified images of original Modi manuscripts, ranging from the era of Maratha ruler Shivaji to the British colonial period, along with their corresponding Devanagari transliterations.

There are an estimated 40 million documents in Modi script across India, including land deeds, scientific manuscripts, and Ayurveda texts.

With very few experts able to read the script today, the AI model addresses a critical gap in heritage preservation.

Built using a Vision-Language Model (VLM) architecture, MoScNet outperforms existing OCR tools and is optimized for use in low-resource environments, where many such manuscripts are stored.

The AI model was developed under the guidance of Prof Sparsh Mittal at IIT Roorkee, with contributions from students Harshal and Tanvi of COEP Pune and Onkar from Vishwakarma Institute.

"MoScNet is light, scalable, and ideal for use even in low-infrastructure regions where these manuscripts are found," said the research team.

MoScNet aligns with major national missions like Digital India, BharatGPT, and Bhashini, and also supports the United Nations’ Sustainable Development Goal 11.4, which focuses on safeguarding cultural heritage.

Researchers believe the technology could be adapted globally to preserve other endangered scripts.

Both the MoScNet model and the MoDeTrans dataset have been open-sourced on Hugging Face, allowing global researchers and developers to build upon the work.