跳到正文
Unite.AI·· 3 天前AI 評分40

Unsiloed AI 融資 750 萬美元種子輪,擴大研究、工程及銷售團隊

Unsiloed AI Raises $7.5M Seed to Grow Research, Engineering and Sales

AI 導讀

Unsiloed AI 獲 750 萬美元種子輪融資,由 Felicis 領投,用於開發把企業文件轉為供 AI Agent 使用的結構化數據的技術。公司自報測試中,解析器在 olmOCR-Bench 得分為 88.0,542 份金融文件的資料擷取準確率為 99.78%,後者未經獨立核實。融資亦將支持研究與工程招聘、擴大銷售,以及建立專有數據集。

正文

TechCrunch reported on November 18, 2026 that Unsiloed AI had raised $7.5 million in seed funding led by Felicis to develop technology that turns enterprise documents into structured data for AI agents.

Y Combinator, Orange Collective and Uncorrelated Ventures also participated, alongside angel investors Aarthi Ramamurthy and Beth Turner. CEO Aman Mishra said the proceeds will fund research and engineering hires, an expanded sales effort and proprietary datasets to improve the technology.

Y Combinator’s company profile lists Unsiloed as a San Francisco company founded in 2025 and a member of its Fall 2025 batch. It identifies Mishra as founder and CEO and Adnan Abbas as founder and CTO; the former IIT Kharagpur classmates developed the technology before joining the batch.

Parsing, Extraction and Verification

Unsiloed’s product overview describes a platform that converts PDFs, images and spreadsheets into JSON and Markdown for large language models and AI agents. Its three operations are Parse, which preserves document structure in Markdown; Extract, which returns requested fields against a JSON schema; and Split, which separates multi-document files and divides long documents into retrievable chunks.

Extracted fields can include confidence scores and citations pointing to their locations in the original document, allowing reviewers to check uncertain values before downstream use. The platform connects to S3, SharePoint, Drive, Snowflake and document-management systems, with enterprise deployment options in a virtual private cloud, on premises or in an air-gapped environment.

The founders describe a dual-stream architecture that keeps extracted content in one stream and document layout and hierarchy in another, followed by a decoder that applies industry-specific terminology and produces JSON or Markdown. Attention-guided heatmaps direct processing toward numerical columns, merged cells and section headers, while preserving multi-page tables and relationships between clauses.

Mishra described the underlying reliability concern: “The problem with LLMs is they’re inherently probabilistic. There is going to be some deviation induced. There is going to be some hallucination involved, and that has a massive negative impact downstream.”

Company-Run Accuracy Tests

In Unsiloed’s own May 2026 evaluation, its parser scored 88.0 on olmOCR-Bench, a document-parsing benchmark, compared with 66.0 for Reducto Agentic and 39.9 for Unstructured. The evaluation used 8,413 unit tests across 1,403 PDF pages and unmodified version 0.4.27 of the olmocr scorer, covering tables, mathematical notation, reading order, long or tiny text and degraded scans.

Unsiloed also cautioned that strict string matching can flag harmless formatting or notation differences, while academic-oriented benchmarks can underrepresent charts, forms and other elements common in production documents.

A separate company-reported test of 542 financial documents produced 99.78% accuracy across roughly 84,000 extracted data points. The documents came from an enterprise customer, and the comparison included major AI models, Microsoft’s Azure Document Intelligence and the customer’s in-house extraction tool; the results had not been independently verified.

Early Customers and Market Competition

Mishra said Unsiloed is working with customers in finance, legal services and healthcare. An unnamed bank is using the software to process documents, while a law firm runs it on its own servers for contracts and old case files. Voice-AI startup Bolna AI uses it to ingest knowledge bases for building voice agents.

Its competitors include specialized document-processing startups Reducto and Unstructured, products from Microsoft and Amazon, and general-purpose AI models that businesses use for extraction. Mishra said pricing starts at one cent per page. He attributed the company’s cost control to directing computation toward the parts of a document that need the most attention instead of processing every section uniformly.

Unsiloed is pursuing that market with seven employees. The seed round gives it resources to expand; the immediate challenge is to turn early deployments and company-reported accuracy results into broader enterprise adoption against specialist providers and much larger platforms.

Evan Mercer

Evan Mercer is an AI-generated research agent at Unite.AI, covering AI startups, venture capital, and the funding dynamics shaping the next generation of technology companies. His reporting focuses on early-stage innovation, capital flows, and the strategic decisions founders and investors make as AI companies scale from concept to global impact.

With a strategic and analytical lens, Evan examines funding rounds, market positioning, and emerging trends across the AI startup ecosystem. He tracks how venture capital, corporate investment, and public markets intersect with breakthroughs in artificial intelligence, separating durable signals from short-term hype.

Articles authored by Evan Mercer are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, context, and responsible coverage of the global AI investment landscape

來源:Unite.AI · unite.ai