Raunak Chowdhuri

Raunak Chowdhuri is co-founder and CTO of Reducto, a document-processing platform that turns PDFs, scans, and spreadsheets into structured, citation-ready data. He studied computer science at the Massachusetts Institute of Technology, and his Y Combinator profile says he published computer vision papers with more than 100 citations before finishing high school.

A Freshman at the Front of a Graduate ML Class

First Round Review tells how he met his co-founder, Adit Abraham. Abraham, then an MIT junior, walked into the first day of a graduate-level machine learning course and found a freshman standing at the front of the lecture hall. He recalls: “The professor introduced Raunak to the class and told us he would be walking us through the first problem set. It was kind of obscene to be sitting there and see this person who had been doing ML research since he was 12.” The two became friends and joined the same living group at MIT.

Rebutting a Viral Claim That GPT-4 Aced MIT

In June 2023, Chowdhuri, Neil Deshmukh, and David Koplow, all MIT electrical engineering and computer science seniors in the class of 2024, published a joint rebuttal titled “No, GPT4 can’t ace MIT.” They found the paper behind the headline “GPT-4 scores 100% on MIT EECS Curriculum” on the morning of Friday, June 16, after it went viral on Twitter. Within two hours, they wrote, they had realized that the dataset itself was suspect.

Their document reports leakage of solution information into the prompts and problems with how the model’s answers were graded. Its Conclusions section invites readers to examine the dataset themselves: “Affirmation can only come through peer evaluation.” The section later reads: “Language models cannot yet be treated as ground truth-generating oracles.”

Hackathons, Remembrall, and the Turn Toward Documents

According to First Round Review, Chowdhuri had wrapped up his previous company and was close to graduating when he and Abraham started entering hackathons together. They won Anthropic’s hackathon, then applied to Y Combinator with an idea drawn from a past hackathon project of Chowdhuri’s and were accepted into the Winter 2024 batch.

Their first product was Remembrall, a long-term memory product for language models. Abraham says customers kept asking for more: “One of the most common feature requests for Remembrall was, ‘Hey, you’re managing my user’s chat history, can you manage the files that they’re uploading as well?’” That request pointed the company toward document processing.

Launching Reducto with Adit Abraham

On January 31, 2024, Chowdhuri and Abraham introduced Reducto in a joint Launch YC post: “We met four years ago while studying computer science at MIT, and we’ve spent the past few years building ML products at companies like Google and NVIDIA. We started building Reducto after struggling with document ingestion while consulting for teams integrating LLMs into their applications.”

Reducto’s about page describes an agentic document platform, lists Harvey, Scale AI, and Vanta among its users, and says it has processed more than 5 billion pages.