dcyphr
Foundergiant: Hello! Tell us a bit about yourself.
Jeffrey Ma: Hi, my name is Jeffrey Ma. I am the founder, CEO, and CTO of dcyphr. I am currently a rising junior at Yale University studying computer science and mathematics. I have previously conducted research in molecular biology, computational neuroscience, and clinical diagnostics.
FG: Introduce your company and the creators of dcyphr.
JM: dcyphr is a crowd-sourced platform informed by AI that makes academic research more accessible. We empower users to request, create, and engage with distillations of research articles. Our distillation network extends globally and consists of students and researchers from various university campuses.
FG: Tell us about your product.
JM: There are two core problems that contribute to the inaccessibility of research:
- Paywall. The money charged by journals does not even go back to supporting researchers. dcyphr makes it free for the average person to engage with research discoveries.
- Complexity. Decoding academic research articles requires time and labor. This process of decoding can either happen once on the writer’s side or N times for N readers on the reader’s side. The status quo opts for the latter -- in doing so, choosing an O(N) solution over an O(1) solution. dcyphr takes advantage of the Internet to cultivate a platform for the O(1) solution.
- The public disengages from science and research because it is inaccessible, contributing to widespread scientific misinformation (highlighted by the COVID-19 pandemic).
- Students, especially from under-resourced backgrounds, are deterred from breaking into STEM.
- Researchers spend a large amount of time sifting through literature that could instead be spent making discoveries.
- Users can curate the platform by requesting research that they wish to see distilled.
- Our NLP model (currently in development) generates an automatic distillation for the user.
- That output is further pruned by other human distillers to be published on our website and to feed back into our model.
FG: How did you come up with the product idea? What inspired you?
JM: I took some biology classes at the beginning of sophomore year that required us to read scientific articles. Even though I had done biological research for 2 years prior, I found this to be an extremely frustrating and tedious activity since the articles failed to communicate the story and concepts of the experiment in a jargon-free manner. None of the students understood the articles, and we ended up spending our discussion sections distilling and summarizing the article. Wouldn’t it be great if someone distilled the article, and we could actually understand it ourselves so that in section, we could instead spend time discussing real experimental concepts and limitations? I did some preliminary research and found that this is a pretty big gap in the status quo that is not being filled (i.e. the problem has been widely studied and reported on and is not just limited to students but extends to the general public, researchers, etc.). Motivated by that, I built the first prototype of the web application, seeded it with a few distillations for the articles assigned in biology classes, and distributed it to those students.
FG: How is your product different from your competitors? Why should clients choose your platform?
JM: We don’t have many competitors. Our few competitors (PaperDigest and Scholarcy) do lack a few things:
- They don’t publish their summaries of articles to be freely browsable, so they don’t solve the paywall problem. If their clients don’t have access to the whole article, they can’t access the summaries.
- They do extractive summarization, and we focus on article distillation. They aim to decrease the amount of information. We aim to reframe the information using psycholinguistic guidelines so that it is easier to read and understand. They also don’t solve the problem of readability and jargon because they merely pull out the top ten relevant sentences from an article to generate a summary. We have humans that distill articles and prune NLP output to avoid those drawbacks. We are also specifically training our NLP model to do this task of distillation rather than summarization.
FG: Was dcyphr bootstrapped or did it receive outside funding?
JM: dcyphr was mainly bootstrapped with a few small non-equity grants.
FG: What is the pricing model of dcyphr?
JM: Right now, we are focusing on developing the product to its full potential. Eventually, we hope to make it a freemium model where the free plan is enough for the casual citizen interested in research, and the premium plan is geared towards institutions to provide a resource for their students and/or researchers.
FG: What advice would you give to budding entrepreneurs?
JM:
- You found the company, not the other way around. In other words, don’t define yourself based on the success or failure of your company.
- Approach it as the best learning opportunity you will ever get. You need a firm grasp of all aspects of your company. Otherwise, someone who does will do it better. If you are a technical founder, you will need to learn and understand the business side of your company. If you are a non-technical founder, you will need to learn and understand the tech side of your company. But more importantly, you will learn how to fail correctly.
FG: Any plans for the future?
JM: We plan to build out the full NLP feature by September. We are also launching a joint research fellowship program with Coronavirus Visualization Team for undergraduates taking gap years due to the pandemic. With humans working alongside the NLP model, we hope to reach 10k distillations by January.
FG: Where can people learn more about dcyphr?
JM:
Check out our linktree: https://linktr.ee/dcyphr
Visit our website: https://www.dcyphr.org
Join our Slack workspace: https://bit.ly/2Bv473O
Sign up for our newsletter for weekly research distillations: https://bit.ly/3ikRclT