# Unsiloed AI brings OCR into modern age

> Source: <https://www.blocksandfiles.com/ai-ml/2026/11/20/unsiloed-ai-brings-ocr-into-modern-age/5311273>
> Published: 2026-11-20 12:36:00+00:00

**Recommendation:** Benchmark document parsers on olmOCR-Bench's roughly 1,400 PDF pages and 7,000-plus binary unit tests, and score with Ai2's unmodified deterministic scorer so results stay comparable across systems.
More in the playbook "Evals and testing for LLM apps": <https://wpnews.pro/playbooks/evals-testing>

AI/ML

# Unsiloed AI brings OCR into modern age

Startup [Unsiloed AI](https://www.unsiloed.ai/) is bringing document scanning into the AI era by updating optical character recognition (OCR) to free digital and analogue scanned paper documents from their format lock-in.

It has raised $7.5 million in seed funding to turn restricted format files such as PDFs, scanned forms, presentations, Word docs and images into structured Markdown or JSON, with applications doing this through Unsiloed API calls. It says financial statements, insurance claims, contracts, clinical records and operational forms carry information critical to everyday business. This information is not just text. Which number belongs to which row, which heading governs a clause, and which axis defines a chart all determine what a document means. Converting a file into a plain-text dump loses that context.

Co-Founder and CEO Aman Mishra said: "We believe AI agents will eventually read and write more documents than humans. Those agents do not just need better OCR. They need a dynamic interface to read, update and act on unstructured data. We are building that interface so documents can become as computable as databases.”

Unsoloed’s AI-driven code, with a dual-stream vision architecture, parses these files, processing content and visual structure together so developers can extract fields against custom schemas, classify and split documents, and update text or images while preserving document formatting. Confidence scores and source citations help teams validate extracted information and route uncertain results for review.

It’s not done with 100 percent accuracy but it’s darn close. Co-Founder and CTO Adnan Abbas said: “We combine deterministic computer vision algorithms with dual-stream vision models to build the interface that ranked first among 19 systems on olmOCR-Bench for AI agents to interact with unstructured data.”

Unsiloed's parser ranked No. 1 among 19 systems in the company's published olmOCR-Bench comparison, scoring 88.0 using Allen AI's unmodified deterministic scorer. The benchmark evaluates document-parsing challenges including complex tables, mathematical expressions, multi-column reading order and scanned pages.

It’s product supports private-cloud, on-premises and air-gapped deployments, enabling customers to process sensitive documents within their own infrastructure.

Unsiloed AI was started up by by Aman Mishra and Adnan Abbas in the 2024-2025 period, and joined [Y Combinator's](https://www.ycombinator.com/) Fall 2025 batch of startups. The seed funding was led by Felicis. Participants include Y Combinator, Orange Collective, Uncorrelated Ventures, Aarthi Ramamurthy,  Beth Turner, and others.

The funding cash will be used to develop Unsiloed's document models, grow its engineering team, and expand cloud and on-premises deployments for enterprise customers

**Bootnote**

The Unsiloed AI

[olmOCR-Bench](https://arxiv.org/abs/2502.18443) suite was created by the Allen Institute for AI (Ai2) to evaluate how well vision-language models (VLMs) and OCR systems convert PDF page images into structured text. It consists of roughly 1,400 PDF pages and over 7,000 binary unit tests that check layout, reading order, math rendering, and table structure.
