Undermind is an AI agent that does a deep and targeted search of the research literature. It finds precisely relevant works, even for complex topics (10-50x better than Google Scholar and Pubmed).
Hi Product Hunt! đ
We are incredibly excited to launch Undermind â 10-50x better search for scientific research.
â The Problem:
Finding the right resources to solve problems is crucial for doing impactful research.
⢠However, researchers often need to find really complex information.
⢠Tools like PubMed and Google Scholar only work for simple phrases. It often takes hours to sift through the (mostly irrelevant) results.
⨠Our Solution:
We rebuilt search from the ground up for complex questions. Our AI agent:
⢠Mimics how a human researcher searches.
⢠Adapts based on what it finds to uncover everything.
⢠Ultimately, is 10-50x better than traditional keyword search engines.
âď¸ How to use it
Just write down your complex research topic, describing exactly what you want, as if to a colleague. After 2-3 minutes, Undermind will give you the precisely relevant articles you need.
đ Why we're doing this
@thomas_hartke and I, the builders of Undermind, are scientists ourselves. Weâve spent a lot of time looking through research literature and have personally experienced those âahaâ moments - when you realize you could have saved 3 months of work if only you knew about that one paper earlier!
Our goal isnât to save you a few minutes of searching yourself; itâs to make it easy to find the solutions that can drastically improve your overall research project.
Just tried it, and I think it works really well.
What I like especially:
* That the system asks for clarifications and then gives an editable summary of what it will search for.
* The clear, detailed and structured presentation of results, with the ability for follow-up questions.
But how did you come up with "10-50x better"? Being a scientist myself, I'm very sceptical about unjustified claims, even if it's "just marketing". And I also think this is highly exaggerated, from experience I know that a careful keyword search can get you very far, and often quite fast also.
@konrad_sx see out whitepaper (link on our homepage) for a quantitative comparison and justification of the huge difference. it comes down to density and comprehensiveness of the results for the typical queries users want to run.
@joshua_ramette Thanks for pointing me to that paper. Stupid of me no to have looked for this myself, but I think you should have mentioned it here (it seems to be common here on PH that people make kind of "10x" claims without any research). Also, I think you should have explained what exactly you mean by "better" here directly.
I just quickly read through the paper, and I see several BIG problems with your method, and I'm sure your conclusions are not supported by the research described. I'll explain this in later when I have time, in detail.
One question I have immediately: In the FAQ you write that currently only abstracts and metadata are searched, while in the paper "searching within the full texts of 2.3 million papers" - ?
@konrad_sx Thanks for your interest and for looking at the details. You're right that the a few comments in the whitepaper are slightly out of date; it was written about our earlier prototype which was initially on just arXiv papers, where we sometimes pulled info from full texts. However, we've found that for the vast majority of topics people search for, the abstracts are sufficient, so when we broadened our database and used the same methods, it should not affect the conclusions much.
Congratulations on the launch! Tried it and it works decently well.
But how do you differentiate from similar tools already in the market with output capabilities that are similar or better? For example, Scispace (typeset.io)?
Been using them for a while and even the free version works well despite the rate limits. Especially like the feature to further deep dive into a specific paper after the literature review gives me a tabular summary.
By the time you build out similar features, wonât such companies already have moved 10 steps ahead? Could be wrong, happy to understand better.
@apm_nerd for complex research topics, the difference is the search quality we can achieve. If you really push our system for specific search results, it shines in a way that fast and direct search systems aren't able to!
@joshua_ramette - this is an amazing idea! This solves my recent biggest frustration with AI: hallucinations. Last time, I thought I found through ChatGPT the perfect research paper for my needs, only to discover that it was completely made up. Will give your product a spin on the free trial and let you know how it goes!
Nice launch guys!
I am into research myself and I have seen this space blow up in the last couple of years. Love the approach taken.
SciSpace still gets my vote in this category on the sheer ease of use and the pace at which it helps my research, plus they also have writing capabilities now.
But excited to see how Undermind develops! Will be watching closely
Hey Joshua,
How does Undermind handle very new or niche research areas where there might be limited published work?
Do you have any plans to incorporate preprint servers or conference proceedings into the search scope?
Congrats on the launch!
@kyrylosilin yes, many preprint servers are already included (like arXiv etc). Our system works particularly well for niche areas -- the adaptive nature lets it dig and find things to hone in right on what you want to know about.
Congrats on the launch of Undermind, @joshua_ramette! đ It sounds like a game changer for researchers struggling with complex queries. Really love the approach of mimicking human search habitsâexcited to see how it reshapes literature reviews. Upvote from me!
This sounds really promising, @joshua_ramette! The idea of mimicking human search behavior for complex queries is intriguing. Iâve often struggled to find relevant research quickly, so I can definitely see the value here. Can you share more insights on how the AI adapts its search strategy? Also, whatâs the pricing model for Undermind? Is there a freemium option for users to test it out first?
@elke_qin yes, we have a free trial and a freemium both! It adapts kind of in the way you do as a human -- looking for new papers based on what's discovered so far and continuing to dig.
Congrats on launching Undermind, @joshua_ramette! đ It sounds like a game changer for researchers drowning in info. Canât wait to see how this shifts the landscape of academic search! 10-50x better?
Congrats on the launch, @joshua_ramette! đ It's about time we had an upgrade in research search tools. The struggle with irrelevant results from PubMed and Google Scholar is all too real. Undermind sounds like a game changer for those complex queriesâexcited to see how it mimics human search behaviors. Itâs like having a research assistant who never runs out of coffee! â Can't wait to try it out and witness the 10-50x magic for myself. Hereâs to saving us all those crushing hours of searching! đŞ #Makers #Launch
I came across this new AI tool from a random ad, it seems it can find complex papers better than Google Scholar, but I'm skeptical if it's truly 10-50x better, as a bank CEO focused on innovation, I'd like to see more transparent data on its performance, also is there a feature for integrating it with existing research library databases?
Interesting idea diving into the deep end of scientific papers but isn't there a risk that relying too much on AI might miss out on some nuanced findings often uncovered in a more manual search, if it's anything like Pubmed or Google Scholar users may see some limitations, I'd like to see some real-life test cases before jumping on board
Undermind