Felicis leads $7.5M bet on Unsiloed AI to make enterprise documents usable by AI agents
Summary
- Unsiloed raised $7.5 million in seed funding led by Felicis to help enterprises make complex documents accessible to AI agents.
- The funding round included participation from Y Combinator, Orange Collective, Uncorrelated Ventures, angel investors Aarthi Ramamurthy, & Beth Turner.
- The startup's software breaks down complex documents containing tables, charts, & complicated layouts to extract structured data for AI agents.
- The technology displays the source location of extracted information within original documents to allow businesses to verify results.
- Co-founders Aman Mishra & Adnan Abbas developed the technology after joining Y Combinator's Fall 2025 batch.
- Customers in finance, legal services, & healthcare are using the software, including a bank, a law firm, & voice AI startup Bolna AI.
- Unsiloed utilizes dual-stream vision models that process text & image data separately to handle complex document elements.
- A heat map-based chunking technique directs computing power toward figures, numbers, & charts to optimize processing.
- The system employs a specialized decoder that accounts for industry-specific terminology to distinguish context-dependent terms.
- A test on 542 financial documents showed 99.78% accuracy across roughly 84,000 extracted data points, though results were not independently verified.
- Unsiloed scored 88.0 on the olmOCR-Bench public benchmark, outperforming Reducto's 66.0 score & Unstructured's 39.9 score.
- Pricing starts at one cent per page, with costs reduced by directing computing resources only to areas requiring attention.
- The San Francisco-headquartered startup currently employs seven people & plans to use the funding to hire researchers, engineers, & expand sales efforts.
With a team of just seven, Unsiloed AI plans to compete with tech giants and specialized startups in the crowded AI document-processing market.
- Keywords
- Startups,AI,Felicis Ventures,Y Combinator,Exclusive,AI agents