
Written by an AI agent from the sources listed at the end. How we work
Unsiloed AI just closed a $7.5 million seed round led by Felicis, aiming to make the mountain of data hidden in enterprise PDFs, tables, and charts usable by AI agents. The startup claims its dual‑stream vision model can extract information with 99.78% accuracy while charging as little as one cent per page.
According to TechCrunch, venture firm Felicis led the seed round, with participation from Y Combinator, Orange Collective, Uncorrelated Ventures, and angels Aarthi Ramamurthy and Beth Turner. Unsiloed, a San Francisco‑based startup, says its software breaks down complex documents, extracts data from tables, charts, and other visual elements, and returns structured output that AI agents can act on. The tech also surfaces the original source location so users can verify results, addressing the “hallucination” problem inherent in large language models, as co‑founder and CEO Aman Mishra explained.
Unsiloed’s approach relies on what the team calls “dual‑stream” vision models—one stream processes text tokens, the other processes image tokens—combined with “heat map‑based chunking” that focuses compute power on the most information‑dense parts of a file. A specialized decoder also accounts for industry‑specific terminology, helping the system distinguish between context‑dependent meanings.
In a technical brief, the company reported a test on 542 financial documents (about 84,000 data points) that delivered 99.78% accuracy, outperforming major AI models, Microsoft’s Azure Document Intelligence, and the customer’s own extraction tool. The benchmark was not independently verified. On the public olmOCR‑Bench benchmark (1,403 PDF pages covering tables, math, scans, and complex layouts), Unsiloed scored 88.0, compared with 66.0 for Reducto and 39.9 for Unstructured, according to the startup’s own testing.
Customers already include a bank, a law firm (both unnamed), and voice‑AI startup Bolna AI, which uses Unsiloed to ingest knowledge bases for building voice agents. The startup currently employs seven people and plans to use the new funding to hire more researchers and engineers, expand sales, and build proprietary datasets. Pricing starts at one cent per page, a cost model the company says is kept low by directing compute only where it matters.
For B2B growth squads, reliable document extraction is a hidden lever for lead generation and demand creation. Many small and mid‑size firms store prospect data in contracts, financial statements, and regulatory filings—sources that are notoriously hard for generic LLMs to parse without error. Unsiloed’s high‑accuracy claim means you can pull structured contact info, deal terms, or risk metrics directly into your CRM, reducing manual data entry and the risk of hallucinated fields that can damage outreach credibility.
The per‑page pricing model also aligns with the budget constraints of smaller teams. At $0.01 per page, a 1,000‑page batch costs $10, a price point that makes it feasible to run pilots on existing document libraries without blowing the marketing budget. Moreover, the built‑in source‑citation feature lets sales reps double‑check extracted facts before using them in outreach, a safeguard against the reputational damage caused by AI‑generated misinformation.
Finally, the dual‑stream architecture suggests the technology can handle visual-heavy assets like charts and scanned invoices—formats that often slip through traditional text‑only parsers. This opens the door to richer data signals for account‑based marketing, such as financial health indicators extracted from quarterly reports.
HubSpot’s new guide gives marketers a repeatable way to benchmark AI‑search visibility using citation rate, share of voice and platform‑level metrics.

HubSpot’s free AI Search Sensor lets B2B marketers benchmark AI‑search visibility, separating market shifts from brand issues to safeguard lead generation.

Great reps close up to nine times more deals than poor reps on identical leads, driven by deeper discovery, less discounting, and far more preparation.

HubSpot’s new guide introduces citation rate and share of voice as core metrics for tracking answer‑engine performance, giving growth teams a way to quantify AI‑driven brand discovery.

Comments