Startups & Funding

Unsiloed AI Raises $7.5 Million to Structure Data for AI Agents

Unsiloed AI raised $7.5 million to turn complex enterprise documents into structured, traceable data for AI agents. Felicis led the seed round as the seven-person startup reported 99.78% test accuracy.

Felicis leads $7.5M bet on Unsiloed AI to make enterprise documents usable by AI agents
Felicis leads $7.5M bet on Unsiloed AI to make enterprise documents usable by AI agentsRodDIYguy / pixabay
By Elena Vasquez6 min read

Why it matters

  • Unsiloed AI raised $7.5 million in seed funding led by Felicis; Y Combinator, Orange Collective and Uncorrelated Ventures also invested.
  • Unsiloed reported 99.78% accuracy across roughly 84,000 data points from 542 financial documents.
  • On startup-run olmOCR-Bench tests, Unsiloed scored 88.0, versus 66.0 for Reducto’s agentic system and 39.9 for Unstructured.
  • Pricing starts at one cent per page, and Unsiloed currently has seven employees.
  • The co-founders joined Y Combinator’s Fall 2025 batch; Unsiloed works with finance, legal-services and healthcare customers.

By Elena Vasquez

Unsiloed AI raised $7.5 million in seed funding led by Felicis, according to TechCrunch, as the startup targets a costly enterprise AI failure: extracting the wrong information from complex documents.

The seed round also included Y Combinator, Orange Collective and Uncorrelated Ventures, TechCrunch reported. Angel investors Aarthi Ramamurthy and Beth Turner participated in the financing, according to TechCrunch.

During an interview with TechCrunch, Unsiloed co-founder and CEO Aman Mishra said: “The problem with LLMs is they’re inherently probabilistic.” He continued: “There is going to be some deviation induced. There is going to be some hallucination involved, and that has a massive negative impact downstream.”

TechCrunch framed the risk in operating terms. Business documents often contain tables, charts and complicated layouts that AI systems can struggle to interpret accurately. The report said errors can become costly and the impact could be severe for banks, hospitals and law firms that rely on these documents for daily work.

TechCrunch’s examples put the commercial stakes inside daily operations, where one bad extraction can become an expensive mistake. Unsiloed’s pitch is to clean the underlying data before an AI agent acts on it.

Who is behind Unsiloed?

Mishra co-founded Unsiloed with Adnan Abbas, his former classmate at the Indian Institute of Technology (IIT) Kharagpur, TechCrunch reported. The two founders joined Y Combinator’s Fall 2025 batch after spending time developing the technology, according to TechCrunch. Unsiloed is headquartered in San Francisco, and it currently has seven employees, Mishra told TechCrunch.

The small team now has fresh capital for a problem that demands both software and domain-specific data work. Its immediate challenge is turning its document parser into dependable infrastructure for businesses where an incorrect answer can trigger costly mistakes.

What is Unsiloed trying to fix?

As businesses increasingly turn to AI agents to automate work, TechCrunch reported, much of the information those agents need remains locked inside complex documents. That creates a gap between a model’s general ability and a bank’s, hospital’s or law firm’s need for dependable document data.

Unsiloed’s software breaks down complex documents, extracts information from tables, charts and other elements, and converts the material into structured data that AI agents can use, the startup told TechCrunch.

The system also shows where each piece of extracted information came from in the original document, Unsiloed said. That mapping lets a business check the result rather than take an AI output at face value, according to the startup.

The distinction is practical: a business can compare a returned figure with the table cell that produced it, rather than accepting the figure without context.

How does its parser work?

Unsiloed uses what it calls “dual-stream” vision models, which process text and image data separately, Mishra told TechCrunch. The startup also uses a technique called “heat map-based chunking” to identify the most important parts of a document, according to Mishra.

“In one of the streams you have the text tokens, in the other stream you have the exact image tokens,” Mishra said. “This is something that we have spent a lot of time architecting.”

Unsiloed then directs more computing power toward figures, numbers and charts, the startup said. The company also uses a specialized decoder that accounts for industry-specific terminology, according to TechCrunch.

Mishra said the decoder helps distinguish between terms that can carry different meanings in different business contexts. The company argues that this vocabulary awareness makes its pipeline more useful for sectors such as finance, legal services and healthcare.

What do the tests show?

A technical brief shared with TechCrunch says Unsiloed achieved 99.78% accuracy across roughly 84,000 extracted data points. The test used 542 financial documents supplied by an enterprise customer, according to the brief.

Unsiloed compared its technology with major AI models, Microsoft’s Azure Document Intelligence and the customer’s in-house extraction tool, the startup said. TechCrunch noted that the results had not been independently verified.

Unsiloed also reported outperforming competing document-processing tools on olmOCR-Bench, a public benchmark developed by Allen AI. In startup-run tests covering 1,403 PDF pages, Unsiloed scored 88.0, compared with 66.0 for Reducto’s agentic system and 39.9 for Unstructured.

The benchmark covered:

  • Tables
  • Mathematical expressions
  • Scanned documents
  • Complex layouts

Because Unsiloed ran the olmOCR-Bench comparison, its lead over Reducto and Unstructured remains a company-reported result rather than an independently checked verdict. The 99.78% internal accuracy figure carries the same limitation, according to TechCrunch.

Who is using the technology?

Unsiloed already works with customers in finance, legal services and healthcare, Mishra told TechCrunch. One unnamed bank uses the technology to process documents, he said.

Another unnamed law firm deployed Unsiloed on its own servers to handle contracts and old case files, Mishra said. The deployment gives the startup a live test for document-heavy legal work while keeping processing inside the customer’s infrastructure.

Bolna AI, a voice AI startup, is a named customer, Mishra told TechCrunch. It uses Unsiloed to ingest knowledge bases for building voice agents, he said.

These deployments span banks, lawyers and voice software. Each case puts the same requirement on the product: preserve the meaning and location of information after complex layouts have been converted for machine use.

How does Unsiloed compete?

Unsiloed operates in a crowded market, TechCrunch reported. The company competes with document-processing startups Reducto and Unstructured, as well as offerings from Microsoft and Amazon.

Businesses also turn to general-purpose AI models to extract information from documents, according to TechCrunch. Those models broaden the competitive set because customers can choose a general system instead of a specialized parser.

Against those alternatives, Mishra argued that Unsiloed’s advantage lies in how its technology processes complex documents. The dual-stream and heat map-based chunking approach adds structure before a model acts on the extracted content.

Pricing starts at one cent per page, Mishra said. He asserted that Unsiloed’s approach also keeps costs down by directing computing power toward document sections that need the most attention instead of processing every section in the same way.

What will the $7.5 million fund?

Unsiloed plans to use the $7.5 million seed round to hire more researchers and engineers, expand its sales efforts and develop proprietary datasets, Mishra told TechCrunch. The company will use the data work to improve its technology, according to Mishra.

The dataset work is aimed at improving the technology, while the hiring and sales plans put more researchers and commercial staff behind it. The next test for Unsiloed is whether its seven-person team can turn startup-run benchmarks and early deployments into broader, independently verifiable performance at scale. Until then, the 99.78% internal accuracy and 88.0 benchmark score remain company-reported rather than independently validated results.

Source: techcrunch.com

Share this article:

More from Elena Vasquez

Elena Vasquez

Show full bio

Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.

708 articles

Related articles

  1. Unsiloed AI Raises $7.5M Seed to Turn Enterprise Documents Into AI-Ready Data
  2. Anthropic's IPO Prospectus Warns Investors of 'Existential Risks'
  3. Mirendil Raises $1 Billion Series A on a One-Researcher Model Bet
  4. Vijil CEO Vin Sharma: AI Agents Need Trust Infrastructure Before Autonomy
  5. One in three AI vendors carries a critical unpatched flaw

« Previous articleNext article »