Felicis leads $7.5M bet on Unsiloed AI to make enterprise documents usable by AI agents

Unsiloed AI raises $7.5 M led by Felicis to turn complex enterprise documents into structured data that AI agents can reliably use.

Felicis leads $7.5M bet on Unsiloed AI to make enterprise documents usable by AI agents

Introduction

Unsiloed AI, a San‑Francisco‑based startup, announced a $7.5 million seed round led by Felicis Ventures, with participation from Y Combinator, Orange Collective, Uncorrelated Ventures, and angel investors Aarthi Ramamurthy and Beth Turner. The funding will accelerate the company’s mission to make the wealth of information locked inside tables, charts, and other complex document elements usable by AI agents across finance, legal, healthcare, and other enterprise sectors.

The problem with today’s document‑AI pipelines

Enterprises rely on PDFs, spreadsheets, and scanned reports for daily decision‑making. While large language models (LLMs) excel at generating text, they struggle with the precise extraction of structured data from visually rich documents. Errors in interpreting a financial table or a legal clause can lead to costly downstream mistakes, especially in regulated industries such as banking, hospitals, and law firms. Existing solutions—general‑purpose AI models, cloud‑based document services, and niche startups—often treat the entire page uniformly, wasting compute on irrelevant sections and still producing hallucinations.

Unsiloed’s technical approach

Unsiloed tackles the challenge with a “dual‑stream” architecture that processes text tokens and image tokens separately. A heat‑map‑based chunking algorithm first identifies high‑value regions—figures, numbers, and complex layouts—so that computational effort is focused where it matters most. A specialized decoder then incorporates industry‑specific terminology, helping the system distinguish between context‑dependent meanings (e.g., “interest” in finance vs. “interest” in legal contracts).

In a technical brief shared by the company, a test on 542 financial documents (about 84 000 data points) reported 99.78 % extraction accuracy, outperforming Microsoft’s Azure Document Intelligence and the customer’s in‑house tool. Unsiloed also claimed a leading score of 88.0 on the public olmOCR‑Bench benchmark, compared with 66.0 for Reducto and 39.9 for Unstructured. While these figures have not been independently verified, they illustrate the startup’s focus on precision and cost‑efficiency.

Early traction and market positioning

Unsiloed is already piloting its technology with customers in finance, legal services, and healthcare. One unnamed bank uses the platform to process internal documents, while a law firm has deployed it on‑premises for contract and case‑file analysis. Voice‑AI startup Bolna AI also leverages Unsiloed to ingest knowledge bases for building conversational agents.

The market for document‑processing AI is crowded, featuring both specialized startups and major cloud providers. Unsiloed differentiates itself by combining visual‑aware models with domain‑aware decoding, and by pricing the service at roughly one cent per page—a model that reflects its selective compute strategy.

Use of the new capital

With seven employees at the time of the announcement, Unsiloed plans to allocate the seed funding toward three core areas:

  1. Talent acquisition – hiring additional researchers and engineers to refine the dual‑stream models and expand the proprietary dataset pipeline.
  2. Sales and go‑to‑market – building a dedicated sales team to target enterprise accounts in regulated sectors.
  3. Data and product development – creating curated datasets that capture industry‑specific document formats, further improving accuracy and reducing hallucination risk.

Practical takeaways

  • For enterprises: Unsiloed offers a way to unlock structured data from legacy PDFs and scanned reports without fully trusting black‑box LLM outputs.
  • For AI developers: The dual‑stream and heat‑map techniques illustrate a promising direction for handling multimodal document content.
  • For investors: The seed round signals confidence in niche AI infrastructure that addresses real‑world compliance and accuracy challenges.

--- Unsiloed AI’s $7.5 M seed round underscores a growing demand for reliable, cost‑effective document extraction that can feed AI agents with trustworthy data. As enterprises continue to automate knowledge work, solutions that combine visual precision with domain awareness are likely to become essential components of the AI stack.

Source

Read original source

Why we picked this

Reports funding for an AI document-processing startup, relevant AI business news.

← Back to all articles