i
News
News · 2026-11-20

Felicis leads $7.5 million seed round for Unsiloed AI

Dataist Dataist

Unsiloed AI has raised $7.5 million in seed funding in a round led by Felicis. The company is developing software that extracts data from complex documents — tables, charts and pages with challenging layouts — and converts it into a structured format for AI agents. The system also identifies exactly where each piece of information came from, so companies can verify the results. Y Combinator, Orange Collective, Uncorrelated Ventures, and angel investors Aarti Ramamurthy and Beth Turner also participated in the round.

Cover: Felicis leads $7.5 million seed round for Unsiloed AI

How Unsiloed processes documents

Business documents often spread data across tables, charts and text. AI systems can struggle to extract it accurately, and mistakes can be costly for companies. The stakes are especially high for banks, hospitals and law firms that use these documents in their day-to-day work.

Unsiloed breaks documents into parts, extracts information from tables, charts and other elements, and produces structured data for AI agents. The system keeps links to the original locations in each document, allowing employees to check the extracted information rather than take the model’s answer on trust.

Unsiloed co-founder and CEO Aman Mishra told TechCrunch that large language models work probabilistically, so they can stray from the correct answer and hallucinate. He said such errors can lead to serious consequences in subsequent stages of work.

Mishra founded Unsiloed with Adnan Abbas, a former classmate at the Indian Institute of Technology Kharagpur. After developing the technology, they joined Y Combinator’s fall 2025 batch.

Early customers and competitors

Unsiloed is already working with customers in finance, legal services and healthcare. One bank uses it to process documents, while a law firm has deployed the system on its own servers to work with contracts and old court cases. Mishra did not disclose the customers’ names.

Among the customers he named is Bolna AI, a startup developing voice AI systems. It uses Unsiloed to upload knowledge bases that it then uses to build voice agents.

The San Francisco company is entering a competitive market. Specialized document-processing software developers include Reducto and Unstructured; Microsoft and Amazon also offer their own products. Companies can also extract information using general-purpose AI models.

Technology and test results

Mishra says Unsiloed’s advantage lies in its ability to process complex documents. The company uses custom “dual-stream” computer vision models: one stream handles text, while the other processes images. Heatmap-based segmentation helps identify the most important parts of a document.

The system then directs more computing resources to elements such as figures, numbers and charts. A specialized decoder accounts for industry terminology and helps distinguish words whose meaning changes with the business context.

Mishra shared results from a test on 542 financial documents provided by an enterprise customer. According to a technical description shared by the startup, Unsiloed achieved 99.78% accuracy across approximately 84,000 extracted data points. The test compared the system with large AI models, Microsoft Azure Document Intelligence and the customer’s internal tool. The results have not been independently verified.

Unsiloed also claims to outperform competitors on olmOCR-Bench, an open benchmark developed by Allen AI. In the startup’s own tests, its system scored 88.0; Reducto’s agentic system scored 66.0, and Unstructured scored 39.9. The benchmark included 1,403 PDF pages and tested the processing of tables, mathematical expressions, scans and complex layouts.

Pricing and plans

Mishra says Unsiloed’s pricing starts at one cent per page. He says selective processing can reduce costs by directing computing resources primarily to the parts that need more attention.

The startup currently has seven employees. It plans to use the new investment to hire researchers and engineers, expand sales and build proprietary datasets to improve its technology.

For Unsiloed, extraction accuracy and the ability to verify data sources are key selling points in a market where customers can choose between specialized products and general-purpose models. But the figures cited are based on the startup’s tests and customer data, and have not been independently verified. Dataist’s take: this limits the ability to compare the claimed accuracy with competitors’ results in real-world conditions.

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X