KEY POINTS
- Unsiloed AI raised $7.5 million in a seed round led by Felicis, with Y Combinator, Orange Collective and Uncorrelated Ventures participating.
- Its software extracts data from tables and charts and shows where each piece came from in the source document.
- The accuracy and benchmark results cited were produced by the startup and have not been independently verified.
Unsiloed AI, a startup that helps AI agents work with complex business documents, has raised $7.5 million in seed funding. TechCrunch reported that the round was led by Silicon Valley venture firm Felicis.
Y Combinator, Orange Collective and Uncorrelated Ventures also took part, along with angel investors Aarthi Ramamurthy and Beth Turner. The company is headquartered in San Francisco and currently has seven employees.
The problem Unsiloed targets is that enterprise documents often contain tables, charts and intricate layouts that AI systems can misread. Mistakes can be costly, especially for organizations such as banks, hospitals and law firms that depend on such files every day.
The company's software breaks documents into parts, pulls information from tables, charts and other elements, and converts it into structured data that agents can use. It also indicates where each extracted item appeared in the original, so users can check results instead of trusting the output blindly.
Co-founder and CEO Aman Mishra said large language models are probabilistic by nature, meaning some deviation and hallucination is unavoidable, and that this can have a major downstream impact. Mishra founded the company with Adnan Abbas, a former classmate at the Indian Institute of Technology Kharagpur. They joined Y Combinator's Fall 2025 batch.
According to Mishra, customers include organizations in finance, legal services and healthcare. One bank uses the technology, and a law firm runs it on its own servers for contracts and old case files, though neither was named. Voice AI startup Bolna AI was named as a customer that uses it to ingest knowledge bases for voice agents.
The market is crowded. Rivals include document-processing startups Reducto and Unstructured, as well as products from Microsoft and Amazon, and general-purpose AI models that companies sometimes use for extraction. Mishra said Unsiloed's edge is a dual-stream vision approach that handles text and image data separately, plus a method called heat map-based chunking that finds the most important parts of a page and directs more computing power there.
The startup shared a technical brief describing a test on 542 financial documents from an enterprise customer, in which it reported 99.78% accuracy across roughly 84,000 data points. It also said it scored 88.0 on the olmOCR-Bench benchmark from Allen AI, against 66.0 for Reducto's agentic system and 39.9 for Unstructured. Those benchmark tests were run by the startup, and the results have not been independently verified.
Pricing starts at one cent per page, Mishra said. The company plans to use the new funding to hire more researchers and engineers, expand sales and build proprietary datasets to improve its technology.