Key facts
- Unsiloed said its software extracts tables, charts and other elements from documents and shows where each extracted item came from.
- Co-founder and CEO Aman Mishra said Unsiloed is working with customers in finance, legal services and healthcare, without naming the bank or law firm.
- Mishra named voice AI startup Bolna AI as a customer that uses Unsiloed to ingest knowledge bases for voice agents.
- Unsiloed reported 99.78% accuracy across roughly 84,000 data points from 542 financial documents in a test it shared; the results were not independently verified.
- In tests the startup ran on the public olmOCR-Bench, Unsiloed scored 88.0 against 66.0 for Reducto and 39.9 for Unstructured.
- Mishra said pricing starts at one cent per page and the seven-person startup will use the funding to hire researchers and engineers.
Unsiloed AI has raised $7.5 million in seed funding led by Silicon Valley venture firm Felicis, the company said in a report published on November 18, 2026. Y Combinator, Orange Collective and Uncorrelated Ventures participated, along with angel investors Aarthi Ramamurthy and Beth Turner.
The San Francisco startup sells software that breaks down complex business documents, extracts information from tables, charts and other elements, and converts it into structured data that AI agents can use. It also shows where extracted information came from in the original document, so businesses can check results rather than take AI output at face value.
"The problem with LLMs is they're inherently probabilistic," co-founder and CEO Aman Mishra said in an interview with TechCrunch. He said deviation and hallucination in model output can have a negative impact downstream. Mishra co-founded Unsiloed with Adnan Abbas, his former classmate at the Indian Institute of Technology Kharagpur; the two joined Y Combinator's Fall 2025 batch.
Mishra said Unsiloed is already working with customers in finance, legal services and healthcare. One bank uses the technology to process documents and a law firm has deployed it on its own servers for contracts and old case files, he said, without naming either. He named voice AI startup Bolna AI as a customer that uses Unsiloed to ingest knowledge bases for voice agents.
Mishra said Unsiloed's approach relies on what it calls dual-stream vision models, which handle text and image data separately, and heat map-based chunking, which directs more computing power toward figures, numbers and charts. A specialized decoder accounts for industry-specific terminology, he said.
Mishra pointed to a test involving 542 financial documents from an enterprise customer in which Unsiloed reported 99.78% accuracy across roughly 84,000 extracted data points, per a technical brief the startup shared. The test compared Unsiloed with major AI models, Microsoft's Azure Document Intelligence and the customer's in-house tool; the results have not been independently verified. Unsiloed also said it outperformed competitors on olmOCR-Bench, a public benchmark developed by Allen AI, scoring 88.0 in its own tests against 66.0 for Reducto's agentic system and 39.9 for Unstructured across 1,403 PDF pages.
Mishra said pricing starts at one cent per page. Unsiloed has seven employees and plans to use the funding to hire researchers and engineers, expand sales and develop proprietary datasets.
