Friday 20 November 2026Updated 02:01 IST
bharat.digital

India's digital economy, public infrastructure and security

Originals · Startup · Startups & Fintech

Unsiloed AI, founded by two IIT Kharagpur classmates, raises $7.5 Mn to make messy business documents readable for AI agents

Felicis led the seed round in the Y Combinator backed startup, which turns PDFs, tables and charts into structured data and shows where every extracted value came from.

Unsiloed AI, founded by two IIT Kharagpur classmates, raises $7.5 Mn to make messy business documents readable for AI agents
Cover: bharat.digital

Unsiloed AI has raised $7.5 million in seed funding led by Silicon Valley firm Felicis, TechCrunch and SiliconANGLE reported on 18 November 2026. Y Combinator, Orange Collective and Uncorrelated Ventures joined the round, along with angel investors Aarthi Ramamurthy and Beth Turner. The startup is headquartered in San Francisco and was co-founded by chief executive Aman Mishra and chief technology officer Adnan Abbas, who were classmates at IIT Kharagpur and went through Y Combinator's Fall 2025 batch.

The problem it is going after is familiar to anyone who has fed a bank statement or a contract into a chatbot. Business documents are full of tables, charts and odd layouts that AI models misread, and in a bank, hospital or law firm a misread number can be costly. Unsiloed's software breaks a document apart, pulls information out of tables, charts and text, and converts it into Markdown or JSON that AI agents can use. It also points back to where each value sits in the original file, so a person can check the output instead of trusting it blindly. "The problem with LLMs is they're inherently probabilistic," Mishra told TechCrunch.

The company says its edge comes from what it calls dual stream vision models, which process text and image information separately, plus heat map based chunking that spends more computing power on figures, numbers and charts. In the company's own tests its parser scored 88.0 on olmOCR-Bench, the Allen Institute for AI benchmark, ahead of rival document tools, and it reported 99.78% accuracy across roughly 84,000 data points from 542 financial documents supplied by one enterprise customer. TechCrunch noted that these results have not been independently verified.

Mishra said customers already include a bank and a law firm that runs the software on its own servers for contracts and old case files, neither of them named, as well as voice AI startup Bolna AI, which uses it to ingest knowledge bases for voice agents. Pricing starts at one cent per page. The team is just seven people today, and the new money will go into hiring researchers and engineers, building out sales and creating proprietary datasets. It enters a crowded field that includes Reducto, Unstructured and document tools from Microsoft and Amazon.

More originals