Essayz
Web Application
Embedded Files
1. Overview
An AI-powered web application for exploring the wisdom of great minds. People can ask for the thoughts of experienced publishers on startups, entrepreneurship, and life from their best essays.
Project Objective: By indexing and analysing ideas from their write-ups, create a simple but user-centric design that allows entrepreneurs to ask questions to publishers, and get instant answers from an AI clone of the publisher's opinion – starting with 2 publishers, Paul Graham and Naval
Scope of Work
User Research
UX Design
UI Design
Prototyping
2. Understanding the Problem
Have you ever wanted to know what your favourite essayist thinks about specific topics without having to search through their publications to find the answer? Or, struggled to find the information you needed from your favourite essays after scrolling through endless pages on Google?
I virtually conducted research interviews to uncover pain points and get more information about the behavioural pattern of how they would approach a simple search task. I worked with 5 people familiar with the works of Paul Graham and Naval, ages ranging between 18-40 years.
My research goal encompassed:
Identifying problems with the user's current approach and their needs
Uncovering opportunities to improve the current approach
Gathering insights and learning about user behaviours and preferences
3. Discoveries from Research
The task
Ask specific search engines for Paul Graham's opinion on "Where do good ideas come from?" and compare the result to Paul's actual opinion on the same topic using a template provided.
The result
The results validated the need to design a platform that answers questions based on the author's opinion on topics, rather than just the generic "whole internet" opinion that search engines or GPT comes back with.
80% of the users reported issues with finding the accurate result, a lot of made-up opinions mixed with facts
62% of the users reported issues with searching for long periods to find an answer that is from Paul Graham's perspective
67% of the users reported they needed to read through a few of Paul Graham's publications to get an answer
66% of the users reported not knowing the direct source of their answers and cannot trace back their steps
90% of the users reported that the experience was not seamless
The solution
The results from the research revealed the challenges users had with the task which was in line with our research goal to identify problems with the user's current approach and their needs, uncover opportunities to improve the current approach, gather insights and learn about user behaviours and preferences.
I grouped the pain points under common themes and potential features in the platform:
Ease of search
Informative navigation
Result accuracy
Result sources
The conclusion
While most entrepreneurs don't have the luxury or access to ask questions directly to Paul Graham, some spend a lot of time browsing through his publications and various sources to understand the key takeaways – an AI clone with his opinion proves to be an affordable alternative.
4. Crafting an Intuitive Technical Solution
From the research, we were able to derive functionalities/models to prioritise for how the platform should work beyond what the user sees
Indexing Essays:
The content from Paul Graham and Naval's essays is ingested
The essays are preprocessed into chunks of text, ranging from sentence to paragraph size
Each chunk is encoded into a vector representation
The vector representations of the chunks are indexed in Postgres using the pgvector extension
Retriever Model:
The question asked is encoded into a vector representation
The question vector is compared against the embedded vectors of the essays to find the knowledge-based text with the closest semantic meaning to the question.
Use of cosine similarity to retrieve the chunks containing meanings most closely related to the question – a mathematical concept used to find the similarity or the relationship between two vectors
Generator Model:
The question and relevant content are fed into a text generation model, specifically GPT-3
Other text generation models that can also be used include BERT, BLOOM etc.
The model uses the question and context to generate an answer
This approach ensures that the generated content is relevant and does not contain made-up information
5. The Design Solution
To achieve the objective of this project, I used insights from the user research to create an end-to-end visual solution:
Low-fidelity wireframes
High-fidelity UI
Interactive design
Prototyping and testing
Tools
Softwares used
Miro
Figma
Notion
Illustrator
Google docs
Google forms
Brand guidelines
Colours, Logo, Typography...
To ensure a consistent and delightful user experience, I harmonized design elements, colors, and typography to reflect the app's personality in every interaction.
Colours
Wireframes
Low-fidelity design
I designed wireframes as they proved useful for identifying areas for design improvements before going into a higher-fidelity UI
First home page iteration
When testing 67% of users reported that they get distracted by the other elements on the sidebar
They found the search was somewhat hidden and would like for it to be more prominent
Second home page iteration
This version was redesigned to have reduced distractions with more focus on the search input
I also introduced an animated informative message, to point users to the search input
Other wireframes ↓
High-fidelity UI
Final design solutions for MVP
Home
The design is tailored to emphasise the search functionality and to improve the user experience when doing so. I also introduced quick actions for users – what the platform is, how it works, and ideas on how to get started in the form of suggested questions
Solution: Seamless experience when navigating the search input and quick actions as guide.
Informative elements
To get the best experience for the platform, I introduced a tool to allow users to select the author they want opinions from, a focus animation pointing users to the search input, a loader with informative text and a comprehensive write-up about how the platform works.
Solution: Informative experience for the user, I introduced elements that helps users get the best experience from the platform
Search result & references
The design is tailored to validate the approach taken – users can see an answer that is relevant to their exact question and does not contain made-up information
Solution: Users get a more specific answer to their questions using the models from the technical solutions, they also see details of where the answer was sourced from
Clickable prototype
Immerse yourself in a seamless user experience designed to captivate and simplify. From intuitive navigation to stunning visuals, this prototype showcases the power of thoughtful design in action – witness user centric interfaces come to life.
Enjoy the journey!
Essayz - Clickable Prototype.mov6. Next Steps
The mission is to reach a point at which AI, whether GPT 3.5 or later can make links between things we’ve never considered in different fields and provide tests we can perform to check its ideas the moment we reach a singularity. Following the launch of the MVP we are looking to introduce some additional features -
Iterative Testing & Improvement: Continuously test and improve the product for an enhanced user experience
August 2023 Launch: Prepare for the exciting launch of the project
Usability testing & feedback: Gather post-launch feedback for further iteration and improvement
Expand authors: Include more essayists/authors to diversify perspectives, currently, we have Paul Graham and Naval
Introduce "reply as author" prompt: Introduce an immediate prompt to reply as a selected author for a more immersive experience
Personalized user inputs: Enable users to input their own essay links, they will get responses tailored to the opinion of the author(s)
Include papers from other authors: Integrate scientific papers and other studies for a comprehensive knowledge base
Develop chatbot: Create a chatbot around stored essays to initiate interactive conversations with authors
These next steps will drive the project forward to deliver an exceptional AI-driven experience.
Next project:
Gum Technologies
Google Sites
Report abuse