Nobel

I always enjoyed reading articles and research papers but found it tough to understand and look for useful sources. So my friends and I built Nobel, an AI powered research assistant in the form of a Chrome extension.

Role

Co-Founder, Head of Product Design

Timeline

Feb - Mar, 2023

Skills

User research, Product design, Interaction design

An AI Tool Without Prompts

We didn't like the trend of AI prompting tools that spoon-fed users with answers. Prompting is terrible UX and since most don't know how to prompt well, they'll never get the best answers.

So, how do we fill in gaps of knowledge to prevent excessive confusion and search fatigue, while incentivizing users to learn and research independently?

User Research
From PhDs to VCs to college students

Our team interviewed 32 people across various disciplines and backgrounds. We learned about what they look for in their research, how they learn, and their frustrations (there were plenty). Here are some of the most notable insights:

Josh Leeman
Materials Research at Princeton
Denver Yu,
Nathania Zhong,
Psychology Research at NYU
The Problem
Synthesizing the problem

User Research Takeaways

1.

It's time consuming to find relevant sources

2.

Users spend a lot of time trying to look for additional sources to fill gaps in their knowledge

3.

Insights users look for evolves. Once users roughly understand a topic, they look for specific insights (e.g. Methodologies, direction, etc.). Overtime, smaller parts of papers are useful

User Problem

Research is a cycle of trial and error searching and reading, where you hope to find some insight that is not only relevant, but also fits your current level of understanding.

The Process
We dislike prompting with a passion

Design Goal

Develop a tool that fills gaps of knowledge by clarifying information and guiding them to relevant sources. It should support users in reaching the level of understanding they need and encourage them to learn independently.

Extension vs. Web

Chrome extensions fit with our goals better than websites because they are present in the research journey -- browsing, reading, etc.

Constraints & Scope

We wanted to launch fast, in 14 days so designs had to service users and the development timeline. Here's our initial feature sets:

* We built in public so check out our updates list

Design Partners

In addition to 32 user interviews, we had 3 design partners, allowing us to observe user behavior on newly prototyped features and experiment as I designed.

The Solution
Your personal research assistant

Feature #1 / Navigation Widget

Enhanced Search

Takes a users’ search and suggests keywords, papers, and better search queries

Summarize Page

Summarize article in simplified terms

Deep Dive

Takes the article or highlighted section and suggests related articles and papers

Explain & Rephrase

Explains highlighted section (gray if nothing is selected)

[1]

In case it's blocking any text, you can move the navigation widget

[2]

Here's what the navigation widget looks like in action. It expands upon hover, revealing a Close button

Feature #2 / Enhanced Search

[3]

Enhanced Search gives you the most relevant keywords you might not know, with definitions, recommended papers, and search queries

[4]

We're built into your journey: enhance your search at the start or mid-Google search!

Feature #3 / Summarize Page

[5]

Summarize pages to see how relevant it is to your research

Feature #4 / Explain & Rephrase

[6]

As research evolves, users look for smaller details. This feature gives more granularity over summaries. Yet, those details can be hard to understand so we'll simplify it for them

Feature #5 / Deep Dive

[7]

Want to get deeper on something? Deep dive on an article or highlighted section! Here, we use boxes to divide sections and create a visual distinction from the Enhanced Search modal

Feature #6 / Interaction Design

[8]

Explain Selection defaults to gray and un-clickable until a user highlights a section

[9]

We made opening, expanding, and moving the extension super satisfying!

[10]

To indicate that feedback has been submitted, it first loads then turns into a check mark icon

[11]

We wanted the website to welcome you into this journey while having that same satisfaction. So, we used rhythm to achieve that. Not bad for only 30 minutes of design and development work

* All pictures contain accurate results and representations of the product

Retrospective
Lessons from 32 Daily Active Users

Google Enters the Market

We stopped working on Nobel due to increasing competitors who had great distribution -- Google Bard, Arc Browser, etc. When they launched months later, it was validating to see that our designs paralleled theirs; they have similar features and designs (especially the UX) from AI in search, highlight and rephrase, etc.

Lessons

I always assumed that people would prefer competitors like PDF.ai because it was more convenient to be spoon-fed with answers. However, our launch taught me that different user types lead to very different optimal features.

Instacart Recipes

Redesigning Recipes (KP Design Fellow Finalist)