As companies more and more flip to AI brokers to automate work, the huge quantities of data locked in complicated paperwork stay a problem. Unsiloed AI has raised $7.5 million in seed funding led by Felicis to deal with that drawback, aiming to assist enterprises make their paperwork extra accessible to AI brokers by means of specialised expertise.
Silicon Valley enterprise capital agency Felicis led the spherical, with participation from Y Combinator, Orange Collective, and Uncorrelated Ventures. Angel buyers Aarthi Ramamurthy and Beth Turner additionally participated.
Enterprise paperwork usually embrace tables, charts, and complex layouts that AI programs can battle to interpret precisely. However getting the knowledge fallacious can result in expensive errors for companies, and the impression might be extreme within the case of a financial institution, hospital, or a legislation agency, as they depend on these paperwork for his or her every day work.
Unsiloed goals to resolve this drawback with software program that breaks down complicated paperwork, extracts data from tables, charts and different parts, and turns it into structured information that AI brokers can use. The expertise additionally reveals the place the extracted data got here from within the authentic doc, so companies can verify the outcomes somewhat than take the AI’s output at face worth.
“The issue with LLMs is that they’re inherently probabilistic,” mentioned Unsiloed co-founder and CEO Aman Mishra (pictured above, proper) in an interview. “There may be going to be some deviation induced. There may be going to be some hallucination concerned, and that has a large damaging impression downstream.”
Mishra co-founded Unsiloed with Adnan Abbas (pictured above, left), his former classmate on the Indian Institute of Expertise (IIT) Kharagpur. The 2 joined Y Combinator’s Fall 2025 batch after spending a while growing the expertise.
Unsiloed is already working with prospects in finance, authorized providers, and healthcare, Mishra advised TechCrunch. One financial institution is utilizing its expertise to course of paperwork, whereas a legislation agency has deployed it by itself servers to deal with contracts and previous case recordsdata, he mentioned, with out naming both buyer. Mishra, nevertheless, named voice AI startup Bolna AI as a buyer, saying it makes use of Unsiloed to ingest data bases for constructing voice brokers.
The San Francisco-headquartered startup has entered a market that’s already crowded and consists of specialised document-processing startups corresponding to Reducto and Unstructured, in addition to choices from Microsoft and Amazon. Companies additionally typically use general-purpose AI fashions to extract data from paperwork, giving Unsiloed one other set of opponents.
Mishra advised TechCrunch that Unsiloed’s benefit lies in the way it processes complicated paperwork. The startup makes use of what it calls “dual-stream” imaginative and prescient fashions, which deal with textual content and picture information individually, together with a way referred to as “warmth map-based chunking” that helps establish a very powerful elements of a doc.
“In one of many streams you could have the textual content tokens, within the different stream you could have the precise picture tokens,” Mishra mentioned. “That is one thing that we now have spent quite a lot of time architecting.”
The expertise then directs extra computing energy towards areas corresponding to figures, numbers, and charts. Unsiloed additionally makes use of a specialised decoder that takes industry-specific terminology into consideration, serving to it distinguish between phrases that may imply various things in several enterprise contexts.
Supporting these claims, Mishra pointed to a check involving 542 monetary paperwork equipped by an enterprise buyer. Unsiloed achieved 99.78% accuracy throughout roughly 84,000 extracted information factors, per a technical transient shared by the startup. The check in contrast Unsiloed with main AI fashions, Microsoft’s Azure Doc Intelligence, and the shopper’s in-house extraction software. The outcomes have, nevertheless, not been independently verified.
Unsiloed additionally claims to have outperformed competing AI document-processing instruments on olmOCR-Bench, a public benchmark developed by Allen AI. In exams performed by the startup, Unsiloed scored 88.0, in contrast with 66.0 for Reducto’s agentic system and 39.9 for Unstructured. The benchmark coated 1,403 PDF pages and examined how properly the programs dealt with tables, mathematical expressions, scanned paperwork, and sophisticated layouts.
Unsiloed’s pricing, Mishra mentioned, begins at one cent per web page. He asserted that the startup’s strategy additionally helps preserve its prices down by directing computing energy towards the elements of a doc that want probably the most consideration, as a substitute of processing each part in the identical approach.
At the moment, Unsiloed has a headcount of seven individuals. The startup, Mishra mentioned, plans to make use of the contemporary funding to rent extra researchers and engineers, increase its gross sales efforts and develop proprietary datasets to enhance its expertise.
Whenever you buy by means of hyperlinks in our articles, we might earn a small fee. This doesn’t have an effect on our editorial independence.