thnkr
An agentic learning and exploration platform that lets anyone harness the benefits of cross-functional diverse intelligence to expand their perspective
Company
n/a
Role
Product designer
Date
November 2024
Tools
Figma, OpenAI
Context
In an era where information is abundant yet often siloed, the ability to access diverse and cross-functional insights is invaluable for personal and professional growth. Recognizing the limitations of traditional learning tools that can inadvertently reinforce echo chambers, I sought to create a solution that fosters expansive and multifaceted understanding. thnkr emerged as an agentic platform designed to empower users to delve into any topic by leveraging a synthetic panel of experts from varied disciplines. This platform not only delivers deep insight but also challenges users' perspectives by presenting conflicting viewpoints, thereby promoting critical thinking and broadening horizons
Objective
My primary objective was to design and build a simple but versatile learning tool that empowers users to explore new topics or critically engage with ones they are already familiar with through diverse expert perspectives. I wanted to avoid homophily (shown below) and help users break out of an echo chamber by providing strong positive versions of conflicting arguments
I didn't define and set a target user for this project as it was a speculative design exploration and my personal goal was to explore how artificial intelligence might help topics more accessible and engaging
Challenges
I had to ensure that the synthetic panel of experts provided genuinely diverse (and even conflicting) viewpoints without overwhelming the user in equivocation
Reading text is not the optimal learning method for everyone. I wanted to make sure I could accommodate users with different learning methods and content preferences (text, audio, and video)
Design process
To get started, I wanted to understand what was possible with solutions that are already available on the market. I tested a variety of platforms for learning, including: Brilliant, Coursera, Perplexity, YouTube, Elicit, ChatGPT, NotebookLM, and Tales.so
If you want to build foundational knowledge, a structured course can be helpful, but requires investing a lot of time
Without careful prompting (and review), LLMs can hallucinate which generate incorrect responses and undermine the credibility of the experience
A lot of learning content is one-dimensional and in order to access more viewpoints you either need to know what they are and search for them or watch longer videos that may or may not be valuable
I conducted several unstructured interviews with friends to understand their learning habits and preferences when encountering unfamiliar topics. I also surveyed which apps they knew about and/or used, if any, when exploring a new topic. The insights I gathered from them highlighted the need for diverse perspectives, the annoyance of unnecessary 'commentary' from LLMs, and the important of multi-modal content delivery
"I want to know what the 'steel man' arguments are for the other side"
"A lot of them [articles] are boring because they treat beginners like idiots"
"I prefer podcasts and audiobooks that I can listen to while walking"
Based on the objective, the desk research, and the interviews, I opened my notebook and sketched out several user journeys that would: a) introduce the platform's concept to users, b) enable them to identify a topic they wanted to explore, and c) consume the generated content (text or audio, I de-prioritised video based on the interviews)
I ended up with a simple object-oriented flow that centred around topics and discussions
After some quick feedback from a friend, I added some fidelity and stitched together a prototype so that I could test the usability of navigating between topics and discussions
After a few rapid iterations on the interface, I began prototyping the system by experimenting with several LLMs and meta-prompts (system outline shown below). I eventually settled on OpenAI to leverage their APIs to take advantage of their text-to-speech generation endpoint
After building out the system, I wanted to reify the concept with a marketing website that demonstrates the concept and the platform's capabilities
Outcome
The design process culminated in a fully realized website for thnkr within a week. The platform generates multiple content formats–text and audio (for now)–allowing the user to engage with topics in their preferred manner. By featuring synthetic panels of experts with diverse and sometimes conflicting perspectives, thnkr effectively broadens users' understanding and encourages critical thinking. While thnkr is in a closed beta, I continue to use it on a daily basis to expand my own perspectives on a wide range of topics
I have left out the development story of this project in the interest of brevity, but it was a joy to work on and allowed me to explore rapid prototyping and development with AI-in-the-loop