Reducto's AI tool turns PDFs and spreadsheets into data LLMs can use. We got an exclusive look at the pitch deck that landed the startup $8.4 million in seed funding.

• Reducto raised $8.4 million in seed funding to develop AI for reading complex documents.
• The startup, part of Y Combinator's winter 2024 batch, aims to improve LLMs' document parsing.
• Investors include First Round Capital, BoxGroup, SVAngel, Liquid2, and notable angel investors.

Large language models struggle to parse data in PDFs and spreadsheets, and one startup just raised a supersized seed funding round to build tech that can read these documents the way humans do.

Reducto, which was founded in 2023 and a member of Y Combinator's winter 2024 batch, is announcing it raised $8.4 million in a seed funding round led by First Round Capital.

Y Combinator, BoxGroup, SVAngel, and Liquid2 also participated in the round, in addition to angel investors Arash Ferdowsi, the founder of Dropbox; Andrew Ofstad, the founder of Airtable; Kulveer Taggar, the founder of Zeus; JJ Fliegelman, the founder of WayUp; Richard Aberman, the founder of WePay, and Ralph Goottee and Tracy Young, the founders of PlanGrid.

There are oodles of data tied up in spreadsheets, jpegs, and PDFs — for example, a healthcare invoice, budget, or a legal document. LLMs often struggle to read them. A PDF could have multiple columns of text, in addition to tables, charts, images, and figures, which are difficult to parse correctly and can lead to hallucinations and output errors.

Reducto has trained multiple models on the various visual cues that matter in documents — from spaces between paragraphs that denote a change in topic to tabs in lists that show nested hierarchies. The goal is to build a generative AI tool that mimics the vision and reading of an actual person, explained Adit Abraham, Reducto's CEO.

"We just finished building what we believe is state of the art for spreadsheet parsing," he told Business Insider. "Our goal across all of this is to be the layer that connects human data with LLMs."

Reducto markets its tech to other companies, with customers including Leya, a legal tech startup that uses the tech to upload legal documents, an AI healthcare company, a late-stage startup that processes documents for US government agencies, and a Big Tech company building out its AI and LLM capabilities. Reducto did not share the names of the latter customers.

There are a slew of productivity-focused startups that analyze and answer questions about PDFs, though many are for personal rather than enterprise use. One, Humata AI, raised $3.5 million from Google's Gradient Ventures a year ago, while OpenAI's ChatGPT Store showcases multiple AI PDF developers.

Check out the 13-slide pitch deck Reducto used to raise $8.4 million in seed funding.