Jain eLibrary turned 4 million scanned pages into JainGPT for 22,000 users

Jain eLibrary began digitizing ancient manuscripts, books, articles, audio files, and scriptures in 2008 because Jain religious texts were scattered across libraries, temple repositories, and specialist organizations. When Anish Visaria joined in 2020, the project had a large but difficult-to-use archive.

Visaria used optical character recognition to extract raw text from more than 4 million scanned pages. Instead of waiting for a full conversational AI system, the team first built Jain Quantum, a Google-like search engine, then launched JainGPT in 2023 so users could ask questions about Jainism and its primary sources.

The staged approach produced measurable reach: Jain Quantum had 790,000 users in the past year, while JainGPT had 22,000 users after launch, most aged 14 to 35. Researchers, professors, diaspora users, and non-Jains also used it, even though the tool was still semantic search rather than a natural-language large language model.

Jain Quantum had 790,000 users in the past year, while JainGPT reached 22,000 users after its 2023 launch, opening primary texts to younger readers and researchers.

Existing Jain texts were difficult to access outside libraries, temple repositories, and specialist organizations.

Younger users needed an easier way to ask questions and receive answers from scriptures and related works.

OCR converted more than 4 million scanned pages into raw text that could support a searchable product.

Building Jain Quantum before JainGPT delivered utility without waiting for a full large language model.

When a specialized archive is hard to use, digitize the underlying corpus first, then add search and conversational access; adoption can grow before the final interface is fully mature.

Jain eLibrary's digital access widened beyond Jain followers: Visaria said JainGPT served researchers and professors as well as users learning rituals, prayers, and Jain philosophy. The tool still needed a natural-language LLM layer, but the archive had already become a searchable, globally usable resource.