Sarvam launches Akshar AI workbench to automate document extraction and proofreading

Digitising documents can be complex. Scanned files, old manuscripts, multi-column reports, and regional language texts often confuse even modern AI tools. Now, Indian AI startup Sarvam is stepping in with a new platform called Sarvam Akshar, which is designed to make document extraction smarter. It is built on top of the company’s Sarvam Vision model to move beyond basic OCR-style text reading and bring reasoning into the process, something many current tools still struggle with. Also Read: Sarvam Kaze Smart Glasses unveiled: India launch timeline, features, and what to expect

Here is everything that you need to know about the Sarvam Akshar.  Also Read: Inside India AI Impact Summit 2026: PM Modi to global CEOs - what happened on day 1

What is Sarvam Akshar?

Sarvam Akshar is a document intelligence workbench that doesn’t just “read” text but tries to understand it. The platform supports English along with 22 Indian languages to show India’s diverse linguistic ecosystem.

How does it work? Instead of extracting words line by line, Akshar analyses the structure of a page, identifying headers, paragraphs, images, charts, and footnotes, so the final output feels organised rather than fragmented. The idea is simple: convert raw scanned files into usable, structured data without forcing humans to spend hours fixing mistakes.

How is it different?

Sarvam says that most legacy Optical Character Recognition (OCR) tools recognise characters but fail to understand layout or context. This often results in broken reading order, especially in newspapers, research papers, or archival material. 

Even newer vision-language AI models, while more advanced, sometimes produce inconsistent outputs and require manual prompt tuning. For industries like healthcare, finance, or government, where accuracy matters, that margin of error is a real concern.

To deal with this, one of Akshar’s standout features is an automated proofreading loop. AI agents flag uncertainties and suggest corrections, while humans can review and approve the changes. According to the company, this could allow experts to validate hundreds of pages in the time it usually takes to manually transcribe one.

Add Techlusive as a Preferred Source

The platform can also caption visual elements like charts and images, making documents more searchable and easier to analyse.