Tech BusinessStart-ups & Funding
Unsiloed AI raises $7.5M seed round led by Felicis
The San Francisco start-up turns complex business documents into structured data for AI agents. It plans to use the seed funding to hire researchers and engineers and expand sales.

Key takeaways
- Unsiloed AI raised $7.5 million in seed funding led by Felicis, with Y Combinator and others participating.
- The start-up converts tables, charts and complex layouts into structured data that AI agents can use.
- It reports 99.78% accuracy in an internal test and a score of 88.0 on olmOCR-Bench, but neither result has been independently verified.
- The company has seven staff and competes with Reducto, Unstructured, Microsoft and Amazon.
In this article
The seed round and its backers
Unsiloed AI has secured $7.5 million in seed funding, with Silicon Valley venture firm Felicis leading the round, according to TechCrunch. Y Combinator, Orange Collective and Uncorrelated Ventures also took part, alongside angel investors Aarthi Ramamurthy and Beth Turner.
The San Francisco-headquartered company says it will put the money into hiring more researchers and engineers, growing its sales team and building proprietary datasets to sharpen its technology. The announcement was reported on 18 November 2026.
Why complex documents are hard for AI
Businesses are increasingly using AI agents, software that can carry out tasks on its own, to automate work. But much of the information those agents need sits inside documents with tables, charts and unusual layouts, which AI systems often misread.
Errors can be costly. A bank, hospital or law firm depends on such documents every day, so a wrong figure pulled from a table can cause serious problems further down the line. Unsiloed's software breaks documents into parts, pulls out data from tables and charts, and turns it into structured information that agents can use. It also shows where each piece of data came from in the original file, so staff can check the result instead of trusting the AI output blindly.
How the technology works
Co-founder and CEO Aman Mishra told TechCrunch that large language models are probabilistic by nature, meaning they can produce deviations and hallucinations, or invented answers, that harm later work. Unsiloed's answer is what it calls dual-stream vision models, which process text and image data in separate streams.
A technique the company calls heat map-based chunking helps it find the most important parts of a page. The system then spends more computing power on figures, numbers and charts. A specialised decoder also accounts for industry-specific terminology, so a term that means different things in different sectors is read in the right context.
Customers and claims that await checking
Unsiloed says it already has customers in finance, legal services and healthcare. Mishra said one bank uses the software to process documents and a law firm runs it on its own servers for contracts and old case files, though neither was named. He did name voice AI start-up Bolna AI as a customer, which uses Unsiloed to prepare knowledge bases for voice agents.
The company points to a test on 542 financial documents from an enterprise customer, in which it reported 99.78% accuracy across about 84,000 extracted data points. It also says it scored 88.0 on olmOCR-Bench, a public benchmark from Allen AI, against 66.0 for Reducto and 39.9 for Unstructured. These results come from the start-up itself and have not been independently verified. Pricing starts at one cent per page, according to Mishra.
A crowded market and a small team

Unsiloed is competing in a busy field. Specialist rivals include Reducto and Unstructured, while Microsoft and Amazon offer their own document-processing services. Some businesses simply use general-purpose AI models for the job.
The company was founded by Mishra and Adnan Abbas, former classmates at the Indian Institute of Technology Kharagpur, and joined Y Combinator's Fall 2025 batch. It currently has seven staff. Whether such a small team can hold its ground against larger rivals and win enterprise trust is the question the new funding will help answer.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



