Pulse Raises $3.9M to Fix Enterprise Data Extraction for AI

Data is the foundation of AI, but the majority of it is in formats that are unsuitable for modelling. Unstructured data, which ranges from financial documents buried in PDFs to contract terms spread across spreadsheets, is a constant bottleneck for businesses looking to build AI-powered solutions. While traditional OCR and AI-based parsing tools have been around for decades, they often fall apart when faced with complex formatting, domain-specific jargon, or intricate tables.

Pulse, a San Francisco-based firm, argues that the current technique to document understanding is fundamentally flawed. Pulse, founded last year by Sid Manchkanti and Ritvik Pandey, has built an API for extracting LLM-ready data from documents while assuring correctness and structure at scale. Nat Friedman and Daniel Gross (NFDG) led the $3.9 million seed round, which also included Y Combinator, Sequoia Capital Scout, Soma Capital, Liquid 2 Ventures, Olive Tree Capital, Tiferes, and executives from NVIDIA, OpenAI, and Ramp.

The funding will allow Pulse to expand its engineering team and extend its extraction capabilities beyond text-based documents to include audio and video, enhancing data ingestion pipelines for enterprises.

Why Enterprises Are Losing 20-30% of Their Critical Data

The problem Pulse is tackling isn’t new. Enterprises have long relied on a mix of legacy OCR tools, manual data entry, and AI models that often fail when faced with real-world complexity. According to Manchkanti, existing solutions lead to data loss rates of 20-30% due to poor extraction, making them unreliable for businesses where precision is non-negotiable—such as finance and healthcare.

“Let’s say you’re a financial institution or a healthcare company. There is no room for an LLM to make something up or hallucinate a number or an error,” Manchkanti said.

Unlike general-purpose AI tools, Pulse’s approach combines intelligent schema mapping with fine-tuned extraction models, ensuring that structured data is preserved without losing context. The company’s technology is already being used across industries, powering workflows for startups and Fortune 100 companies alike.

The Demand for Structured Data in AI

Structured data organized in neat rows and columns is easy for AI models to process. But in the real world, most data is unstructured. According to IDC, 90% of the world’s data falls into this category, encompassing everything from customer contracts to investment reports and sales presentations.

For companies building AI copilots, digital agents, and retrieval-augmented generation (RAG) pipelines, getting high-quality training data is a constant challenge. Many enterprises still rely on human workers to manually extract, clean, and label data, a process that is slow, expensive, and error-prone.

Pulse’s solution automates this process, enabling businesses to convert raw, unstructured data into machine-learning-ready formats with minimal human intervention.

How Pulse is Being Used

Pulse’s platform is already being deployed across a range of industries, helping companies transform unstructured data into structured, AI-ready datasets. Some recent examples include:

• A Fortune 100 enterprise is using Pulse to convert PDFs, images, and spreadsheets into structured datasets for its production RAG pipeline, improving retrieval accuracy.
• A Y Combinator startup has automated its investment workflows by processing complex financial CIMs (Confidential Information Memorandum), reducing due diligence time from weeks to days.
• A public investment firm is extracting and normalizing data from thousands of real estate rent rolls to power an ML-driven market intelligence product.
• A growth-stage startup has eliminated manual data entry in its accounting workflows using Pulse’s schema-enforced extraction pipeline, saving over 2,000 hours monthly.

These use cases highlight the growing demand for reliable unstructured data processing, especially as more enterprises seek to build AI-driven applications with internal data.

Pulse enters a market that has already attracted significant investor interest. Startups like Unstructured have raised $65 million to tackle similar data ingestion problems, while Instabase recently secured $100 million in funding to expand its unstructured data processing toolkit.

However, Pulse is betting that its API-first approach built specifically for software teams will set it apart. Unlike traditional OCR and document parsing tools, Pulse is designed to handle the messiness of real-world documents without requiring extensive manual intervention.

With fresh funding, the company plans to expand its extraction capabilities to handle multimodal formats, including audio and video, allowing enterprises to generate higher-quality training data. The focus for 2025 is to maximize the intelligence extracted from every document, making data ingestion as seamless as possible.

For now, Pulse remains a small but fast-growing team in San Francisco, solving some of the toughest challenges in enterprise data processing. As AI adoption accelerates, companies that can harness their unstructured data effectively will have a significant advantage. Pulse is betting that its approach will be the one that finally delivers on that promise.