The key facts
- Unsiloed AI raised $7.5 million led by Felicis with participation from Y Combinator, Orange Collective and Uncorrelated Ventures
- The technology turns tables, charts and complex layouts into structured data with source references in the original document
- Customers operate in finance, legal services and healthcare; the market competes with Reducto, Unstructured, Microsoft and Amazon
$7.5 Million Seed Round
The startup Unsiloed AI has raised $7.5 million in seed funding, with the round led by Silicon Valley venture firm Felicis. Y Combinator, Orange Collective and Uncorrelated Ventures also participated, with angel investors including Aarti Ramamurthy and Beth Turner. The San Francisco-based company develops software that parses complex corporate documents and turns them into data suitable for AI agents to work with.
The Problem: Documents That AI Understands Inaccurately
Business documents often contain tables, charts and complex layouts that AI systems interpret with errors. For banks, hospitals and law firms that rely on such documents in their daily work, the cost of a mistake can be high. Unsiloed co-founder and CEO Aman Mishra attributes this to the nature of large language models: 'The problem with LLMs is that they are inherently probabilistic. Deviations arise, there are hallucinations, and this has a serious negative downstream impact.'
How the Technology Works
Unsiloed's product breaks down complex documents, extracts data from tables, charts and other elements, and turns it into structured data. A key feature is that the system shows where each fragment in the original document came from, allowing companies to verify the result rather than take the AI's output on faith. Technically, the startup uses 'dual-stream' vision models that process text and images separately, and a 'heatmap-based chunking' method that directs more computing resources to numbers, charts and diagrams. A specialized decoder accounts for industry terminology, distinguishing terms with different meanings in different business contexts.
Claimed Results and Customers
According to the startup's technical report, on a test of 542 financial documents from a corporate client, Unsiloed achieved 99.78% accuracy across approximately 84,000 extracted data points, comparing itself against major AI models, Microsoft Azure Document Intelligence and the client's in-house tool; these results have not been independently verified. On the public olmOCR-Bench benchmark from Allen AI, according to the company's own tests, Unsiloed scored 88.0 versus 66.0 for Reducto's agentic system and 39.9 for Unstructured; the benchmark covered 1,403 PDF pages. Mishra named the voice AI startup Bolna AI as a client; in addition, a bank processes documents through Unsiloed's technology, and a law firm deployed it on its own servers to work with contracts and old cases — the names of these clients are not disclosed.
A Crowded Market and Plans
Unsiloed is entering a saturated market where specialized startups Reducto and Unstructured operate, along with solutions from Microsoft and Amazon; general-purpose AI models that companies use for data extraction also pose competition. Pricing starts at one cent per page. The team currently has seven people; the raised funds will go toward hiring researchers and engineers, expanding sales, and building proprietary datasets. Mishra founded the company together with Adnan Abbas, his classmate from IIT Kharagpur; they went through Y Combinator's fall 2025 batch.


