BharatGen AI pilot reads Sanskrit maths, astronomy and medicine texts

BharatGen, the government-backed artificial intelligence (AI) project based at IIT Bombay, is working with India's Ministry of Culture on a pilot of AI systems that can read, transcribe, search and analyse Sanskrit manuscripts on subjects such as mathematics, astronomy and medicine, according to a column published on Sunday, 8 November 2026. The pilot is part of the ministry's Gyan Bharatam mission, which plans to use AI to read and digitise more than one crore (10 million) old manuscripts, according to reports published on Saturday, 7 November, and 8 November. The plan relies on optical character recognition (OCR, software that turns printed text in an image into editable text), handwritten text recognition (HTR, which does the same for handwriting) and AI tools, with the digitised texts to sit on an AI-powered National Digital Repository that users can search and question by text or voice, the reports said.

India holds one of the world's largest collections of manuscripts, written over many centuries on palm leaves, birch bark and paper, in Sanskrit and many other languages and scripts. Much of this material sits in temples, monasteries, libraries and private homes, often in fragile condition, and far fewer people can read the old scripts today than in the past. Knowledge on medicine, mathematics, astronomy, grammar, philosophy and the arts lies in these texts, but most of it has never been catalogued, let alone edited and published.

Scanning a manuscript only preserves an image. To make the text useful to scholars and students, someone must read the script and type it out, a slow job that needs rare expertise. Handwritten text recognition and AI language models could speed up that step greatly, turning page images into searchable text that can then be checked by human experts. A repository that can be questioned in plain language, by text or voice, would also open these texts to readers who are not specialists.

There are real limits. Old scripts vary from region to region and from scribe to scribe, many pages are damaged, and AI systems can misread characters or make up text. That is why the reported plan pairs AI with expert review, and why the pilot focuses first on specific subjects. For a knowledge-systems project, accuracy matters more than speed: a mistranscribed verse or formula can mislead later readers.

The partnership with BharatGen also matters. BharatGen is a government-funded effort, led from IIT Bombay, to build AI models for Indian languages. Training such models on classical texts could help them handle Sanskrit and older forms of Indian languages better.

Peepals calculation: The government's June 2026 figures of about 1.29 lakh manuscripts publicly accessible against 44 lakh documented mean that only about 2.9% of documented manuscripts were open to the public at that point (1.29 ÷ 44 ≈ 0.029). The 44 lakh documented equals 44% of the mission's one-crore target (44 lakh ÷ 100 lakh = 0.44). Our calculation from the reported figures.

Short-term: The AI pilot on Sanskrit texts in mathematics, astronomy and medicine will test whether machine reading is accurate enough to speed up transcription.

Long-term: If the tools work, far more manuscripts could become searchable online, giving students, researchers and the public access to texts that are now locked away in fragile bundles. It could also create new work for scholars who can check and edit machine transcriptions.

Who is affected: Scholars of Sanskrit and classical Indian languages, libraries, temples and families that hold manuscripts, AI researchers working on Indian languages, and students of Indian knowledge systems.

Questions and Answers