companies

Unstructured

Open-source ETL for turning messy PDFs, slide decks and scanned documents into clean, chunked text that RAG pipelines and LLMs can actually use.

Brian Raymond founded Unstructured in 2022 after years building document-processing systems at Primer and, earlier, at the CIA, where he saw how much of a data scientist's time went into wrangling PDFs, Word files, and scanned forms before any modeling could start. The open-source library he built handles that first step: it partitions more than 60 file types into structured elements — titles, tables, list items, captions — rather than just splitting text by character count. That element-level parsing has made Unstructured a common building block in retrieval-augmented generation pipelines, where clean chunking has a direct effect on retrieval quality. The library has been downloaded millions of times and shows up in thousands of public GitHub repositories, with connectors for pulling source documents from S3, SharePoint, Salesforce, and dozens of other systems. Backed by Bain Capital Ventures, Menlo Ventures, and Madrona among others, Unstructured also sells a managed platform for teams that want the same partitioning and chunking pipeline running at scale without operating it themselves.

Founded

2022

Headquarters

Sacramento, United States

Sector

data