Radically good scientific search

We help you find exactly what you need, no matter how complex. Our AI assistant digs through the literature for you, thoroughly and carefully, like a human researcher.

Trusted and Used by Researchers at
Features

A solution engine

Imagine finding the best solutions to every problem, and always working on the most important goals. That's priceless.

Here's how we help you do it:

Incredibly complex topics

You can describe exactly what you want to Undermind, as if to a colleague.

Accurate results

Undermind finds the precise papers you need, with clear explanations about why they matter.

Comprehensive discovery

Undermind figures out how many papers exist on a topic, so you know how much has been done.

Brainstorm with a copilot

Undermind helps you find the most important questions to ask.

About Us

Next Generation AI-Powered Search Algorithms

Undermind isn't like other search engines. Instead, we reimagined search from the ground up in order to mimic a human's careful, systematic discovery process. During each search, we examine results in stages, and use language models to make key decisions, such as recognizing crucial information and adapting the search strategy. This leads to unprecedented accuracy and comprehensiveness.

We benchmarked the first version of Undermind to be 10-50x better than Google Scholar, and we've been improving ever since.

Learn More
about-Ai-powered-search-image
shape
FAQ

Frequenty Asked Questions

We cover all fields of science using the Semantic Scholar database of over 200 million articles, which contains PubMed, ArXiv, and many other databases.
Each search takes 2-3 minutes to complete because, as our AI agent starts to identify the best papers, it actually adapts and change its search methods, just like you would. This adaptation makes it possible to uncover all the relevant results, but also makes the search slower than a simple Google or PubMed search.

If a search takes less than 30 seconds, it can't possibly be this good.
For now, we search the abstracts and metadata of scientific articles in the Semantic Scholar database. We plan to incorporate full text discovery to allow for even more complex discovery goals.
The simple answer is we've redesigned the search experience from the ground up to achieve the best search quality possible, because we know that's what matters to researchers.

Every other "search engine" you've used really just returns recommendations, not precisely relevant results. If you're asking for something general or something that's common knowledge, that's sufficient, since most of the results will be helpful. However, that's not what scientists need. For researchers, what you need is often so complex that you have to wade through hundreds of results. It's painful.

You should think of Undermind as sending a talented colleague to search for you, who will carefully, thoughtfully filter through papers to find exactly what you need.
If you're looking for a very specific paper, there's a small chance it isn't present in the Semantic Scholar database of over 200 million articles which we draw from.

Another possibility is that the Undermind search has not yet converged. Unlike in a traditional search engine, where all results are returned at once, ordered from best to worst, we perform a search in stages, like you would as you carefully search the literature. We look at a the most promising set of papers first, and depending on the difficulty of finding relevant results within these papers, we may not discover all the relevant papers in this first pass. By statistically modeling this process, we can determine what fraction of relevant papers we have likely found. If your paper didn't show up at first, it's likely the search system knows it has not yet found everything (and the report summary will tell you). You can deploy more computation ("extend" the search) to find the rest. See our whitepaper for additional explanation.
Think of Undermind as a support structure beneath a researcher's intellectual pursuits - like the roots of a tree. Our goal is to support researchers with intellectual tools that truly augment their capabilities as we transition into the age of AI-powered scientific discovery.
Built by researchers

Started by two quantum physics PhDs from MIT, with decades of experience in deep research.

shape
Josh Ramette
CEO
shape
Tom Hartke
CTO