Unsiloed AI: IIT Kharagpur classmates raise $7.5 million seed led by Felicis

Representative image: niklaspatzig / Pixabay
Unsiloed AI, founded by IIT Kharagpur classmates Aman Mishra and Adnan Abbas, raised $7.5 million on 18 November 2026 to build document tools for AI agents.
Unsiloed AI: what happened
Unsiloed AI, a document-processing start-up co-founded by two former classmates from the Indian Institute of Technology (IIT) Kharagpur, announced on Wednesday, 18 November 2026 that it has raised $7.5 million in a seed round led by the venture firm Felicis. Y Combinator, Orange Collective and Uncorrelated Ventures also invested, along with angel investors Aarthi Ramamurthy and Beth Turner. The company says it will spend the money on hiring researchers, engineers and sales staff.
Unsiloed AI: key facts
- Round: $7.5 million seed funding, led by Felicis, announced on 18 November 2026.
- Other investors: Y Combinator, Orange Collective, Uncorrelated Ventures, and angels Aarthi Ramamurthy and Beth Turner.
- Founders: Aman Mishra (co-founder and chief executive) and Adnan Abbas, who studied together at IIT Kharagpur.
- Accelerator: the founders went through Y Combinator's Fall 2025 batch.
- Team size: seven people, according to the company.
- Product: software that turns PDFs, scans and other complex files into structured data that AI agents can read; pricing starts at one cent per page, Mishra said.
- Customers: firms in finance, legal services and healthcare; voice AI start-up Bolna AI is a named customer.
Unsiloed AI: why it matters
For students who want to build rather than only apply for jobs, Unsiloed is a clear example of a path that is open to Indian engineering graduates: two classmates from the same campus, a place in an accelerator, and then an institutional seed round within about a year. An accelerator is a programme that gives a young company a small investment, mentoring and introductions to investors in return for a stake. A seed round is usually the first sizeable money a start-up raises from professional investors, meant to prove the product and find steady customers.
The problem the company works on is a practical one. An AI agent is a software program that can carry out tasks on its own, such as reading a contract or filling a form. Most business knowledge sits in messy files: scanned pages, tables, charts and forms. Basic OCR (optical character recognition, the technology that turns a picture of text into typed text) often loses the layout, so a table can come out as a jumble of numbers. Unsiloed says its software breaks a document into parts, pulls out data from tables and charts, keeps the original layout and links each extracted value back to the place it came from, so a person can check it. Mishra said AI agents will eventually read and write more documents than humans do, and that they need more than basic OCR.
The company's performance figures are its own claims, not independent results. It says its product scored 99.78% accuracy on about 84,000 data points taken from 542 financial documents of an enterprise customer, and 88.0 on olmOCR-Bench, a public benchmark (a standard test) from the Allen Institute for AI, against 66.0 for its rival Reducto and 39.9 for Unstructured. The company itself cautioned that strict string matching in such tests can count harmless formatting differences as errors. The market is crowded: Reducto, Unstructured, Microsoft and Amazon all offer document tools.
Peepals calculation: an accuracy of 99.78% means 0.22% of values were wrong, which on about 84,000 data points is roughly 185 errors (84,000 × 0.0022 ≈ 185), and the test set works out to about 155 data points per document (84,000 ÷ 542 ≈ 155). On olmOCR-Bench, the company's score is about one-third higher than Reducto's (88.0 ÷ 66.0 ≈ 1.33). Our calculation from the reported figures.
| Company | Score on olmOCR-Bench (as reported by Unsiloed) |
|---|---|
| Unsiloed AI | 88.0 |
| Reducto | 66.0 |
| Unstructured | 39.9 |
What should a student reader take from this? First, that investors are paying for narrow, unglamorous problems, such as reading invoices and contracts correctly, when the team can show measurable results. Second, that a small team can win a funded round: Unsiloed has seven people. If you are building something similar on campus, the useful next step is to test your product on real files from a real user and publish how you measured it, as this company did.
Unsiloed AI: impact
Short-term: Unsiloed plans to grow its research, engineering and sales teams with the new money, which means new roles at a seven-person company. The firm has not said where those roles will be based.
Long-term: if more companies hand work to AI agents, tools that read documents reliably become basic infrastructure, much like payment systems did for online shopping. That creates room for more specialised start-ups, including ones founded by Indian engineers.
Who is affected: students and young engineers interested in AI, applied machine learning and enterprise software; banks, law firms and hospitals that handle large volumes of paperwork; and rival document-processing companies.
Unsiloed AI: key takeaway
Treat Unsiloed's accuracy numbers as company claims, and watch whether its hiring and customer list grow over the next year as a sign of whether narrow AI tools can scale.
Unsiloed AI: questions and answers
Who founded Unsiloed AI?
Unsiloed AI was co-founded by Aman Mishra, who is its chief executive, and Adnan Abbas. The two were classmates at IIT Kharagpur and joined Y Combinator's Fall 2025 batch.
What does Unsiloed AI's software do?
It takes complex files such as scanned PDFs, tables, charts and forms and turns them into structured, machine-readable data that AI agents can use. Each extracted value is linked back to its spot in the original file so that a person can verify it.
How much does Unsiloed charge?
Mishra said pricing starts at one cent per page. The company has not given its full price list in the material published with the funding news.
Are Unsiloed's accuracy figures independently verified?
No. The 99.78% accuracy result came from the company's own test on 542 financial documents from one enterprise customer, and the benchmark comparison with Reducto and Unstructured was also published by Unsiloed. Independent tests have not been reported.
Disclaimer: Prepared by the Peepals newsroom from publicly available sources with AI assistance. Information is accurate to the best of our knowledge at the time of publication and may change. Images may be representative. Not professional advice. Report an error via our contact page.
By Rhea Malhotra, Staff WriterCovers Digital Life and more for Peepals Yuva. All stories by Rhea Malhotra
Vaibhav Sooryavanshi, 15, youngest T20 captain; leaves Bihar for India A
Rakshitha Ramraj, 19, beats Chen Yu Fei, exits Hong Kong Open in quarters
Your week ahead, every Sunday. Free.
Exam and result dates, IPOs opening, scheme deadlines, government holidays and the RBI decision for the next 7 days, plus the week's top stories. One email a week, unsubscribe in one click.
Rhea Malhotra is a pen name of the Peepals Global Editorial Team, which researches, fact-checks and publishes every story. The portrait is an illustration.
Responsible for this story: the Peepals Global Editorial Team, led by editor Prashant Sonar. How we report









