Funding

Felicis backs Unsiloed AI with $7.5M to unlock enterprise documents for AI agents

Unsiloed AI has secured $7.5 million in seed funding to help enterprises make complex documents accessible to AI agents, addressing a critical challenge as businesses deploy automation at scale.

·4 min read
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 agents

Enterprise organizations deploying AI agents face a persistent obstacle: critical business information trapped within documents featuring intricate layouts, tables, and charts that automated systems struggle to parse reliably. Unsiloed AI is tackling this challenge with $7.5 million in seed funding led by Felicis Ventures, bringing specialized technology designed to transform document data into structured formats that AI agents can actually use.

The funding round included participation from Y Combinator, Orange Collective, and Uncorrelated Ventures, alongside angel investors Aarthi Ramamurthy and Beth Turner. The San Francisco-based startup joined Y Combinator's Fall 2025 batch after co-founders Aman Mishra and Adnan Abbas spent time refining their approach.

Documents containing visual elements and complex formatting present a fundamental problem for large language models. When extraction goes wrong—whether through hallucination or probabilistic error—the consequences can be severe for financial institutions, healthcare providers, and law firms that depend on document accuracy for daily operations. Unsiloed's platform addresses this by decomposing complex documents, extracting data from visual and textual elements, and converting everything into structured information while maintaining traceability back to source material.

The problem with LLMs is they're inherently probabilistic. There is going to be some deviation induced. There is going to be some hallucination involved, and that has a massive negative impact downstream.

Aman Mishra, Unsiloed co-founder and CEO

Mishra and Abbas, former classmates at the Indian Institute of Technology Kharagpur, have already attracted customers across finance, legal services, and healthcare. A financial institution is processing documents through the platform, while a law firm has deployed it on internal servers to manage contracts and historical case files. Voice AI startup Bolna AI represents another customer, using Unsiloed to build knowledge bases for voice agent development.

The competitive landscape includes established players such as Reducto and Unstructured, alongside enterprise offerings from Microsoft and Amazon. General-purpose AI models also serve as indirect competitors for document extraction work. Unsiloed differentiates through its technical architecture and specialized processing approach.

Technical differentiation

The startup employs what it describes as "dual-stream" vision models that process text and image data through separate pathways, combined with "heat map-based chunking" to identify document sections requiring the most attention. This architecture allows the system to allocate computational resources strategically rather than treating all document content equally.

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

Aman Mishra

The platform incorporates a specialized decoder that recognizes industry-specific terminology, enabling it to distinguish between terms with different meanings across business contexts. This contextual awareness helps prevent misinterpretation of domain-specific language.

Performance metrics

In testing with 542 financial documents from an enterprise customer, Unsiloed achieved 99.78% accuracy across approximately 84,000 extracted data points, according to a technical brief provided by the company. The test compared Unsiloed against major AI models, Microsoft's Azure Document Intelligence, and the customer's internal extraction tool, though these results have not undergone independent verification.

On olmOCR-Bench, a public benchmark developed by Allen AI, Unsiloed scored 88.0 in testing conducted by the startup. This compared favorably to Reducto's agentic system at 66.0 and Unstructured at 39.9. The benchmark evaluated performance across 1,403 PDF pages, testing handling of tables, mathematical expressions, scanned documents, and complex layouts.

Pricing and growth plans

The startup's pricing model begins at one cent per page. Mishra argued that the focused computational approach also reduces operational costs by concentrating processing power on document sections that require it most.

Unsiloed currently operates with seven employees. The company intends to deploy the new funding toward hiring additional researchers and engineers, scaling sales operations, and building proprietary datasets to enhance its technology.