Unsiloed AI raises funding to process enterprise documents


San Francisco-based startup Unsiloed AI has raised $7.5 million in a seed funding round led by Felicis, with participation from Y Combinator, to develop software that extracts structured data from complex business documents for AI agents. The platform focuses on processing tables, charts, and mathematical formulas while preserving original source links to facilitate human review.
To enable AI agents to operate efficiently within enterprises without errors in reading complex tables and files, Unsiloed AI has secured a $7.5 million seed funding round to develop a specialized architecture that turns difficult documents into auditable, structured data.
San Francisco-based Unsiloed AI has raised $7.5 million in a seed funding round led by venture firm Felicis, with participation from accelerator Y Combinator and other investors, according to a report by TechCrunch on November 18, 2026, at 04:00 PST (16:00 GST), also covered by Dealroom.
The startup develops software designed to extract structured information from complex business documents so that AI agents can use it in downstream workflows. General-purpose models often face clear difficulties in accurately processing documents containing tables, charts, mathematical expressions, and unfamiliar layouts.
The company was founded by Aman Mishra and Adnan Abbas, former classmates at the Indian Institute of Technology Kharagpur who joined Y Combinator’s Fall 2025 batch together. Their software relies on separate processing pipelines for both textual and visual information within a single document.
The platform directs extra computing effort toward pages or regions containing complex material, and aims to maintain direct links connecting every extracted value to its original location within the document so that users can inspect and review evidence directly.
The feature of tracking the origin of information is extremely important in enterprise environments; if an AI agent uses an incorrectly extracted number inside a financial model, that error can propagate and compound across subsequent decisions based upon it.
Although the service displaying the precise location of the number within the document makes the human review process easier and faster, the mere presence of a reference pointer to the original page does not in itself guarantee that the underlying interpretation of that data was entirely correct.
Unsiloed AI stated that it works with organizations in the finance, healthcare, and legal sectors, and disclosed Bolna AI among its customers while keeping other customer names unannounced. The startup says that a test conducted with a customer involving 542 financial documents and about 84,000 data points achieved an extraction accuracy of 99.78 percent.
It also announced positive results on olmOCR-Bench, a general evaluation measuring proficiency in handling difficult PDF layouts; however, all of these figures originate from tests conducted by the startup itself and have not been independently reproduced under equivalent conditions.
The company noted that its starting price is about one U.S. cent per page, though this figure does not determine the total cost of running a fully integrated enterprise AI agent, which may also require retrieval, storage, model inference, and human verification services, at a time when the company plans to hire new talent and develop its own proprietary datasets.
Its list of competitors includes established document processing platforms and major general models; although the investment amount is relatively small compared to funding for leading models, the problem it addresses has a fundamental commercial impact because agents require reliable structured data before acting. Upcoming evidence will consist of independent extraction tests, failure rates with unfamiliar documents, data security practices, and actual customer usage volume.
Q: Who led Unsiloed AI’s seed funding round and who are its founders?
A: Felicis led the $7.5 million round with participation from Y Combinator, and the company was founded by Aman Mishra and Adnan Abbas, graduates of IIT Kharagpur.
Q: How does Unsiloed AI’s software process complex document pages?
A: It uses separate processing pipelines for text and visual elements, focusing extra computing power on complex regions while linking extracted values to their original locations in the document.
Q: Has the announced 99.78 percent accuracy rate undergone independent external auditing?
A: No, the 99.78 percent figure derived from 542 financial documents and about 84,000 data points, along with olmOCR-Bench results, is based on the company’s internal testing and has not yet been independently verified.
Q: Does the one-cent-per-page price represent the full cost of running an intelligent agent in enterprises?
A: No, this is only the starting price for page extraction, whereas the total cost for an enterprise agent includes expenses for retrieval, storage, model inference, and human review.