Do not index
Do not index
One of the biggest challenges with searching for information is refining initial queries to improve the relevancy of search results. Generally our first query is pretty bad. We might have a vague sense of the content we’re looking for, but we’re not sure quite how to express it. We might be missing a key term, lacking the right source, or need better phrasing for the question that we have. With nearly infinite content at our disposal, how do we find the perfect needle in the haystack that we’re looking for?
Using AI as a search tool has exploded this year. Yet many challenges still arise with building a great search experience, that re-imagines the search workflow from first principles.
At Epsilon, we’re focused on simplifying how we search academic literature. With scientific research, the problems mentioned above are often more pronounced. Science is dense, filled with jargon, and hyper specific. For those breaking into a new field or conducting interdisciplinary research, it can take weeks of research to stumble upon that one paper that outlines the right term you’ve been looking for. Part of our mission to make science more searchable is to improve this workflow.
To do so, we’ve built several features that we call “Search Branching”. Here are some of the features we’ve incorporated.
Search Branching
Text as a control element
We’re really excited about the idea of using text as a control element. By that we mean allowing users to take actions on AI generated text. As an example, I could be searching for effects of carbon taxes, see a point that I find interesting and click on that bullet. In Epsilon, you’re given the options to elaborate on the point, or view arguments for and against that point.
This offers a great way to expand on content that piques an interest without having to painstakingly search for new information. With a single click, Epsilon starts streaming more relevant content.
In the future we plan to add many more actions for researchers to take. As an example, we think it would be incredibly powerful to click on a point made in the summary and ask our AI to extract raw data to support that claim.
Suggested Follow Ups
Conducting great scientific research starts with having great research questions. Great questions often come from starting with a simple question and adding multiple different perspectives. Acquiring this can come from experience, community, or from lots of reading.
With Epsilon new angles to your research question are automatically generated and displayed underneath initial search queries. Users have told us that this has helped jog lots of interesting ideas that they would not have initially considered.
Citation Exploration
In scientific research one of the common branching workflows once a good paper is found, is to look through that paper’s citations and references. In Epsilon, we’ve made extracting these super easy.
For a search, we show a table of all the papers that were considered when generating the summary. Within each paper, we added an easy way to navigate the citation graph directly from that table. A simple dropdown shows all the relevant citation information that one can explore. Papers can be viewed from here or easily saved for later reference.
Final Takeaways
LLMs have given builders many new tools to build great search experiences. We’re excited to innovate with this tooling in a way that can accelerate science forward. We expect many search products to incorporate and iterate on features like these to simplify the process for finding precise information.
If you’d like to try out Epsilon, you can do so for free here: www.epsilon-ai.com.
