Unsiloed AI raises $7.5 million to help AI agents read company documents
AI agents
Computer helpers that can complete several tasks.
structured data
Information arranged so software can use it more easily.
dual-stream vision models
Models that handle text and image information separately.
What happened
Unsiloed AI, a startup that makes business documents easier for AI agents to use, raised $7.5 million in seed funding. TechCrunch reported the round. Felicis led it. Y Combinator, Orange Collective, and Uncorrelated Ventures joined. Angel investors Aarthi Ramamurthy and Beth Turner also participated.
Why business documents are difficult
Important company knowledge often sits inside documents. These files mix paragraphs, tables, charts, images, and unusual layouts. AI can miss how those pieces connect. It can also read a number incorrectly. Such mistakes can become costly for banks, hospitals, and law firms.
What Unsiloed says it does
The software breaks complex documents into parts. It extracts information from tables, charts, and other elements. Then it turns the information into structured data for AI agents. It also marks the original location of each extracted item. People can therefore check an AI answer against the source document.
CEO Aman Mishra said large language models are probabilistic. He warned that hallucinations can cause serious downstream harm. Unsiloed uses what it calls dual-stream vision models. One stream handles text. The other handles image data. Its heat-map chunking method tries to find important areas. The system sends more computing power to figures, numbers, and charts. A specialized decoder also considers industry-specific terms.
What the performance numbers show
In a test using 542 financial documents from an enterprise customer, Unsiloed reported 99.78% accuracy. The test covered roughly 84,000 extracted data points. It compared the system with major AI models, Microsoft Azure Document Intelligence, and the customer’s internal tool. TechCrunch noted that the results were not independently verified.
Unsiloed also reported its own test on the public olmOCR-Bench. It scored 88.0. Reducto scored 66.0. Unstructured scored 39.9. The test covered 1,403 PDF pages. It examined tables, mathematical expressions, scanned files, and complex layouts. These results are useful signals, but they remain company-run comparisons.
Customers and competition
Unsiloed says it already works with customers in finance, legal services, and healthcare. One unnamed bank uses it for document processing. An unnamed law firm runs it on its own servers for contracts and old case files. Voice AI startup Bolna AI is a named customer. It uses Unsiloed to load knowledge bases for voice agents.
The market is crowded. Reducto and Unstructured focus on document processing. Microsoft and Amazon offer competing tools. Businesses can also use general-purpose AI models. Unsiloed says pricing starts at one cent per page.
What to watch next
The company has seven employees. It plans to use the new money to hire researchers and engineers. It also wants to expand sales and build proprietary datasets. The big unanswered questions are broader than one test. Can independent testers repeat the results? Does accuracy hold across industries and document types? Can the company keep costs low as customers grow?
The next evidence will come from customer deployments, independent evaluations, and clearer performance data. If AI agents take on more business work, traceable document answers may become as important as fluent answers.
(Source: TechCrunch)
Unsiloed AI raises $7.5 million for easier document reading
📰 Full story: Unsiloed AI raises $7.5 million to help AI agents read company documents
A small company wants AI to use messy business documents more safely.
AI agents
Computer helpers that can complete several tasks.
seed funding
Early money used to help a young company grow.
benchmark
A test used to compare different tools.
💡 The gist
- Unsiloed AI, a company making business documents easier for AI, raised $7.5 million.
- It reads words, tables, charts, and page layouts.
- It shows where its information came from.
Companies keep important facts in documents. These files can look messy. They may contain paragraphs, tables, charts, pictures, and unusual designs. AI can read one part correctly but miss how parts connect. A wrong number can cause a wrong business decision. Banks, hospitals, and law firms face this risk.
Unsiloed AI wants to help. AI agents are computer helpers that can complete several tasks. The company turns difficult documents into organized information for those helpers. It also marks the original place for each piece of information. A person can check the answer against the document. This matters because AI can sometimes make up or change facts.
Felicis led Unsiloed AI’s seed funding. Y Combinator and other investors joined. The company says it already works with finance, legal, and health customers. One bank uses it. A law firm runs it on its own servers. Voice AI company Bolna AI also uses it.
The market has strong competitors. They include Reducto, Unstructured, Microsoft, and Amazon. Some businesses also use general AI models. Unsiloed says its price starts at one cent per page.
The company describes special methods. It looks at text and images in separate streams. It gives more computer power to numbers, charts, and figures. It also considers words that have special meanings in different industries.
Unsiloed tested 542 financial documents from one customer. It checked about 84,000 pieces of data. The company reported 99.78% accuracy. TechCrunch says an independent group has not checked this result. Unsiloed also reported better scores in its own test on a public benchmark. Company tests can be useful. They do not prove the tool works everywhere.
Unsiloed has seven employees. It plans to hire researchers and engineers. It will also build more training data and expand sales. The next questions are clear. Can outside testers repeat the results? Will accuracy stay high across many industries? Will the price remain low as usage grows?
(Source: TechCrunch)
A company helping AI read hard papers
📰 Full story: Unsiloed AI raises $7.5 million to help AI agents read company documents
Unsiloed AI helps AI read company papers with tables and pictures.
Unsiloed AI
A company that helps AI read company papers.
What happened
Unsiloed AI is a company that helps AI read company papers. It raised $7.5 million. That money helps the company grow.
Company papers can have words, tables, charts, and pictures. AI may mix those parts up. Then AI can give the wrong answer.
Unsiloed AI makes the information easier to use. It also points to the original paper. A person can check the answer there.
The company tested 542 papers about money. It says 99.78% of its answers were correct. An independent group did not check this test. So we do not know how it works everywhere.
Unsiloed AI has seven people. It plans to hire more people to study and build the tool. It also wants more examples to teach the tool.
The company has customers in money work, law work, and health work. The next step is showing the tool works in real jobs.
(Source: TechCrunch)