This Is How a 26-Year-Old Raised $108M in 1.5 Years | Reducto, Adit Abraham
These notes were generated by AI and may contain inaccuracies.
Did a ton of manual unsexy work in the early days. Tried hiring an initial data labeling team but they weren't accurate enough, so spent a lot of time labeling boxes on documents. If you stacked all the pages that Reducto processed, it would be 10 times the height of Mount Everest back when fully unautomated. Manually set up every single subscription for Stripe billing and setup. These repetitive, boring tasks were okay because the focus was on whether the company was moving forward. Signing up a new customer is a privilege.
Hi, my name is Adit Abraham. I'm the co-founder and CEO of Reducto. Reducto is a platform that helps AI teams parse, extract, and edit complex unstructured data for language model use cases. Reducto has grown incredibly quickly, raised $108M in total funding from investors like a16z (Andreessen Horowitz), Benchmark, and First Round. Powers ingestion for Fortune 10 enterprises and leading AI companies like Harvey, Rogo, and Meru. Processed more than a billion pages and growing every week.
Ever since young, had side hobbies. In high school, saw article about Flappy Bird creator making $50,000 a day on ad revenue. With best friend, decided to make apps, even discussed skipping college. Things didn't work out that way, went to college and pursued longer career. Inspiration showed going off the beaten path can lead to outlier outcomes. Valuable to see individual effort from side projects spiral into something bigger.
That led to going to MIT, undergrad in computer science. Took first grad-level ML course on metalearning. Professor introduced Ronic (freshman, first week on campus) as learning assistant for course primarily PhDs. Became close friends. When Ronic suggested working together, immediate yes without thinking twice about leaving job. Admired him enough for it to be a no-brainer.
Before YC batch, gave up on revenue multiple times. Tested ideas, got to people willing to pay but urgency and need weren't high enough. For Rememberall (first long-term memory API for language models to remember past mentions, store and retrieve context), customers at best willing to pay $50-100/month. Not number one priority, more nice-to-have.
Built file management feature (managing uploaded files like managed RAG service) using off-the-shelf tools. In demos, people got excited about file management. Trained own models, posted technical blog in YC forum on document segmentation via simple Streamlit app (upload document, draw boxes). Customers pulled: "These are better results than my existing vendor. Is this a hosted API? Do you have a Stripe link? Can I purchase?"
Slap in the face how much market wanted the product. High conviction came from solving immediate problems people actively sought solutions for. Long-term memory would be needed eventually, but not immediately.
Customers demonstrate product value not just by money, but time invested to make it great together. Aggregated data to train state-of-the-art models, but production data differs (intensive financial, healthcare, insurance—not on internet). Customers brought esoteric examples: doctor annotations at page bottom relating to top; financial tables with thousands of rows.
Design partner relationships: customers provide feedback, iterate daily. By end of first week, significant progress. They care to make product great, on same team. Set up individual Slack channels with customers, have their phone numbers, call directly. If issues, work late nights because trust from early stage is not taken lightly. Reasons to choose established companies, but pay back trust by being there individually.
Even today, customers can ping Ronic or me directly. Part of Reducto is us as their ingestion team.
Instead of just marketing claims, put product in front of people via public playgrounds, even imperfect. Proved it works on hardest documents. Led to fast growth. As two-person company, trillion-dollar enterprise booked demo after uploading hard documents that failed other vendors.
Wouldn't have engaged if shy about building. When claiming most accurate product, mean it—show examples, let them test themselves. Companies tried selling to 2-3 times hesitated trusting seed-stage, but came back inbound impressed by month-over-month progress. Year 1 struggles now year 2 customers.
Knew investors from building company. Early founders think in firm brands (tier 1 VCs with decades reputation), but key is individual partner for next 10 years—through successes (fundraises, contracts) and bad moments (layoffs, lost contracts, media incidents). See how interactions change when things go poorly.
Liz (seed investor) always available (texts at 10pm get 10:05 reply). During rare time off (Broadway show with husband), had OpenAI outage. Ronic messaged frantically, she stepped out, called network, got chief product officer on phone to help (despite not being important customer).
Advice: Take time to understand investor commitment—most important decision.
Behind exciting moments, unseen effort. First big enterprise contract required on-prem deployment (never done before, no infrastructure engineers). Woke up, went to office until exhausted, slept few hours, repeated. Everyone dove in: engineers did customer support, ops labeled data for ML. No role silos—focus on helping company succeed via product working and happy customers.
Reducto not just parsing: layer connecting human data to new intelligence across all data. Products built with Reducto create net new documents via agentic workflows. Future AI products: foundation model intelligence + context. Reducto aims to be best way to interact with that context, building block for use cases.
Keep EO in your library
Save the videos and channels worth coming back to, and find them again in one place.
