AI 热点

Unsiloed 获 $7.5M 种子轮融资,将扫描文档转为 AI 智能体可用的结构化数据

SiliconANGLE更新于 11月19日 03:40

Unsiloed nabs $7.5M to turn scanned documents into structured data for AI agents - SiliconANGLE

AI 导读

Unsiloed 获得 $7.5 million 种子轮融资,拟将 PDF 等复杂文档转换为 AI 智能体可读的 Markdown 或 JSON。其解析器在 olmOCR-Bench 获 88.0 分,并支持私有云、本地及完全隔离环境部署。

正文

Document-scanning firm Unsiloed Inc. today announced it has raised $7.5 million in seed funding to turn PDFs and other complex documents into structured, readable data for artificial intelligence agents.

Participants in the round, led by Felicis, included Y Combinator, Orange Collective, Uncorrelated Ventures, Aarthi Ramamurthy and Beth Turner.

A variety of documents arrive in formats that are not straight data or text. Financial statements, insurance claims, contracts, clinical records, and operational forms make up the constellation of critical information for everyday business and operations. This information could be stored as images, PDFs or other formats and include charts, layouts and other elements that don’t translate directly into text.

Determining which column or axis of a chart belongs to which row, which heading leads which clause, or whether a pull quote interrupts the flow of the text or appears at the end of the document becomes a question of how to store the information.

Unsiloed parses documents such as PDFs using intelligent visual-first scanning to understand structured tables, sections and parent-child relationships between text. The company, which describes itself as intelligence-directed and vision-first, technology can scan forms, presentations, Word files and images into Markdown or JSON, a JavaScript Object Notation standard format readily readable by large language models.

“We believe AI agents will eventually read and write more documents than humans,” said co-founder and Chief Executive Aman Mishra. “Those agents do not just need better OCR. They need a dynamic interface to read, update and act on unstructured data.”

OCR, or optical character recognition, is the standard technology that is used to convert physical or image-based text, such as scanned paper documents or PDFs, into digital text. However, OCR doesn’t recognize tables, graphics or layouts; it only pre-processes the document to identify text, recognizes text and then outputs a contextual stream based on what it sees.

Unsiloed goes a step further by accounting for the page’s layout and format and providing a structured, machine-readable document that inherits the document’s original features. This means that if the presentation included headings, charts or other parts that relate to one another, those relationships are included in the data.

OCR is the most common way companies digitize documents today; Unsiloed believes it can do it better.

The company’s parser rated an 88.0 score on the olmOCR-Bench benchmark, a benchmark published by the Allen Institute for AI that is a rigorous, reproducible evaluation of how well modern AI models handle real-world OCR challenges. Compared with other services, Unsiloed said, it stacks up extremely well; for example, Anthropic PBC’s Claude Opus 4.7 ranks at 81.9, OpenAI Group PBC’s GPT 5.5 ranks at 84.6 and Google LLC’s Gemini 3 Pro ranks at 77.7.

“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,” said co-founder and Chief Technology Officer Adnan Abbas.

The company said its dual-stream vision architecture allows it to process content and visual layouts at the same time. Developers can upload custom schemas to extract fields, classify and split documents, and update text and images while preserving document formatting.

It said it has processed millions of pages of documents for organizations across industries, from early-stage AI startups to banks, publicly traded insurance companies and law firms. The platform supports private-cloud, on-premises and fully air-gapped machines, allowing enterprise organizations to build Unsiloed into their document pipeline and do processing on their own infrastructure.

“We’ve significantly enhanced accuracy, context understanding, and response quality across all knowledge-base-powered conversations,” said Prateek Sachan, founder and CTO of Bolna AI, a company that builds voice agents for languages spoken across India.

Image: SiliconANGLE/Microsoft Designer

来源

SiliconANGLE