Reducto, the most accurate ingestion platform for unlocking unstructured data for AI pipelines, announced that it has raised a $24.5M series A round of funding led by Benchmark, alongside existing investors First Round Capital, BoxGroup and Y Combinator. This follows the company’s October 2024 seed round of $8.4M and brings its total funding to date to $32.9M.
“Reducto’s unique technology enables companies of all sizes to leverage LLMs across a variety of unstructured data, regardless of scale or complexity,” said Chetan Puttagunta, General Partner at Benchmark. “The team’s incredibly fast execution on product development further underscores their commitment to delivering state-of-the-art software to customers.”
Reducto turns complex documents into accurate LLM-ready inputs, allowing AI teams to reliably use the vast data that’s locked in PDFs and spreadsheets. Ingestion is a core bottleneck for AI teams today because traditional approaches fail to extract and chunk unstructured data accurately. These input errors lead to inaccurate and hallucinated outputs, making LLM applications unreliable for many real-world use cases such as processing medical records and financial statements.
Also Read: Opsera and Lineaje Partner to Deliver Agentic AI-Powered Auto-Remediation, Gold Secure Images, and Open Source Security at Scale
In benchmark studies, Reducto has been proven to be significantly more accurate than legacy providers like AWS, Google and Microsoft – in some cases by a margin of 20+ percent, alongside significant processing speed improvements. This is critical for high-stakes, production AI use cases.
Over the past year, Reducto has processed hundreds of millions of pages of unstructured data for companies like Airtable, Scale, and a Fortune 10 enterprise. Customer Legora reports that with Reducto, its team tripled processing speed and built a pipeline that processes millions of documents per month, all while reducing engineering time spent on chunking by 90%.
Reducto’s founders Adit Abraham and Raunak Chowdhuri met while at MIT and spent several years building machine learning products at companies like Google and Nvidia before striking out on their own. “We worked as ML engineers, and learned early on that often, inaccurate output and hallucinations are not necessarily model problems, but ingestion problems – bad inputs lead to bad outputs,” said Abraham, co-founder and CEO of Reducto. “Nearly 80% of enterprise data is in unstructured formats like PDFs, making it hard to access and analyze. To overcome this, we developed vision models that read documents the way humans do — breaking down complex layouts visually and contextually parsing each region. When customers see Reducto for the first time, they immediately get the significance. We’re grateful to Benchmark and all our investors for sharing our excitement over this breakthrough.”
“Having invested in Reducto from day one, I immediately recognized Adit and Raunak’s technical brilliance and unique approach to solving the ingestion problem,” said angel investor Kulveer Taggar, who participated in Reducto’s seed round. “The fact that leading AI teams are overwhelmingly choosing Reducto over incumbents validates their approach – when your AI’s accuracy depends on perfect inputs, there’s no room for compromise.”
Source: GlobeNewswire