<aside> 👷🏽 We’ve just entered a closed Alpha testing stage! If you are interested in seeing the app at it’s current stage, you can sign up here to book a slot!


If you just want to stay in touch or hear more as we keep developing, join our slack or our discord.


Abstract Poetry asks the question - how can we make academic search meet your needs as an individual researcher better?

When talking to researchers, we found three problems with search as it is:

  1. Keyword-search demands time, or prior expertise and still limits results to those with high-quality metadata
  2. Search algorithms use popularity based metrics which show what most people might want to see, but bury results for anyone who isn’t part of the plurality
  3. Researchers use social connections to stay up to date which reinforces their own view of the world instead of allowing them to truly represent the literature.

All of this creates a bias in search that leads makes it just that much more difficult to achieve any researchers goal. To publish high quality, reproducible, and representative research that adds to the current literature.

This document is a summary of what we have so far - and where we are going.



<aside> 🎓 We’re building interactive, exploratory search for academic papers that considers what you as a researcher want in order to refine search criteria.


The overarching problem with academic search engines now is that while you may not have the exact search keywords, you probably have a broad intuition about the type of things you’re looking for. Your goal isn’t necessary to find a specific paper, but instead generate a reading list or collection of interesting articles related to a field or topic. In short, you’re performing exploratory search, rather than discriminative search.

search relevance is not what u need, when you don’t know, what you don’t know

Existing keyword search engines (Google Scholar, or even the preferable Semantic Scholar) are poorly designed for this type of exploration, for several reasons:

  1. They return the same results each time, and the search doesn’t update based off of what you want to see. This means you have to make many unique searches to find different responses, and often end up scrolling through pages of results
  2. Searching provides functionality for filtering only using simplified metadata such as time, date, publication. When looking for papers, you often also want to be able to discover papers with similar methods, topics, or underlying theories