My Role
UX UI Designer
Responsibilities
Research · Interaction · Visual Design · Testing
Duration
Aug 2020 – Jun 2021
SUS Score
67.6%
Overview
The Challenge with Automated Extraction
Even though there are automated systems to extract information from research papers, we can't confidently verify whether this information is correct or not.
In deep learning, data is the foundation — model performance directly depends on the quality and accuracy of the data you feed in. Yet the manual annotation process that ensures this quality was time-consuming, tedious, and frustratingly error-prone.
"We are moving slowly into an era where big data is the starting point, not the end."
Problem Context
Why Accurate Annotation Matters
Automated systems to extract information from research papers exist, but their accuracy is often questionable. In the context of deep learning, the quality of data directly influences model performance. However, the process of manually annotating data is often time-consuming, tedious, and prone to human error.
Given the importance of accurate data, this process is critical for ensuring that machine learning models perform at their best. The problem wasn't just technical — it was deeply human. We needed to understand the researchers, data scientists, and industry experts who lived inside this process every day.
⚠️
High Error Rate
Manual annotation suffers from inconsistency and human bias
⏱️
Low Efficiency
Tedious workflows drain researcher time and motivation
📉
Model Impact
Poor data quality directly degrades deep learning performance
Discovery
Getting to Know Users
At the start of the project, I supported extensive user research and competitor analysis to familiarize data scientists and project managers with the current annotation products. For me, that stage is crucial — I love to start projects by gathering as much information and data as I can, and analysing them to better understand the problem I'm trying to solve.
Due to the constraints posed by the COVID-19 pandemic, we conducted online surveys and telephone interviews to gather valuable insights into customer needs.
01
Stakeholder Interviews
Telephone and online interviews with data scientists and project managers to surface their workflows, pain points, and existing tool habits.
02
Competitor Analysis
Deep dive into existing annotation tools to identify patterns, gaps, and opportunities for differentiated design.
03
Survey Research
Online surveys to gather quantitative data on user preferences, frequency of annotation, and satisfaction with current tools.
04
Iterative Testing
Low-fidelity paper sketches evolved into high-fidelity mockups, validated through continuous usability testing cycles.
Personas
Who Are We Designing For?
Based on our research, we identified two primary user archetypes whose needs shaped every design decision throughout the project.
Dr. Sarah Müller
Senior Data Scientist
Age 32 · Research Institute
"I spend more time fixing annotation inconsistencies than actually training models. There has to be a better way."
✓
Goals
Ensure high-quality, consistent annotation data for her ML models
Reduce time spent on manual review and error correction
Have full visibility into annotation progress across the team
!
Pain Points
Current tools are unintuitive and require technical onboarding
No feedback loop — she can't tell if annotators understood the task
Context switching between papers, tools, and spreadsheets
James Okafor
ML Project Manager
Age 38 · BioTech Startup
"Keeping the team motivated through repetitive annotation tasks is the hardest part of my job. I need a tool that helps with that."
✓
Goals
Maintain team productivity and motivation over long annotation projects
Track contribution and quality across multiple annotator roles
Deliver labeled datasets on time and within budget
!
Pain Points
No performance tracking — hard to identify top contributors vs. laggards
Annotation quality drops off significantly after the first few weeks
Lack of reward or recognition structures for team effort
Approach
Designing for Delight
Our mission was to reimagine the annotation process for data scientists. Rather than viewing it as a mundane task, we aimed to transform the experience into something that data scientists would enjoy. To achieve this, we worked closely with data scientists to understand their workflows, pain points, and the unique context in which they operate.
This collaboration was challenging but highly rewarding, as it provided us with invaluable insights into how we could improve the tool's usability. By conducting user interviews and iterative testing, we developed an annotation application that was intuitive and efficient.
We focused on creating a streamlined user interface that minimized friction, enhanced productivity, and made the process of annotating research papers feel less like a chore and more like a rewarding task.
MVP Definition
User Roles & Gamification Structure
Based on our sketching sessions and deep collaboration with data scientists, we identified three primary user roles for the gamified annotation application.
🔍
Role 01
Selector
The Selector chooses research papers and categorizes them by Disease, Variant, and Gene, ensuring the papers are correctly assigned for annotation.
✏️
Role 02
Annotator
The Annotator labels relevant information from the papers, extracting key data points and relationships between Disease, Variant, and Gene.
🎯
Role 03
Reviewer
The Reviewer, an industry expert, verifies the accuracy of the data provided by the Selector and Annotator, ensuring high-quality annotations.
Annotation Workflow
Selector
Assigns Papers
→
Annotator
Labels Data
→
Reviewer
Validates Quality
Gamification Features
To motivate users and sustain engagement across long annotation projects, we introduced two core gamification elements:
🏆
Leadership Board
A Leadership Board tracks user points based on their contributions (Selection, Annotation, Review), fostering healthy competition and continuous engagement across the team.
🎁
Earn Rewards
Users earn lunch vouchers as rewards for their efforts in annotation, selection, and review tasks, encouraging ongoing participation and creating a tangible incentive structure.
Design
High Fidelity Mockups
The final tablet designs use a cohesive purple gradient system — transforming annotation from a dry task into an engaging, mission-driven experience across every screen.
1
Splash & Onboarding
Splash Screen
First-launch experience — brand intro with tagline, key stats, and a single CTA to get started
2
Login & Sign Up
Authentication Screen
Clean login with Microsoft SSO and Google Workspace — minimal friction entry to the workspace
3
Dashboard
User Dashboard
Overview of active missions, personal stats, and available tasks to join
4
Mission Setup
Disease Selection
Pick a disease mystery game — Cystic Fibrosis, Alzheimer's, Breast Cancer and more
Mission Target
Select a specific research target — CFTR Trafficking with player counts and mission details
5
Role Selection
Choose Your Difficulty Level
Select Selector, Annotator, or Reviewer role — each with distinct skills and difficulty
6
Core Annotation Flow
Calibration Check
Before annotating — verify you're reading the correct paper with a calibration question
Full Paper Reading
Read the full research paper with highlighted entities before annotating
Assay Method
Structured annotation of assay methods — answer targeted questions about experimental procedures
Protein Activity Annotation
Answer structured questions about protein functions extracted from the paper
7
Rewards & Leadership Board
Mission Congratulations
Mission complete — earn points, view mission progress summary, and continue to next question
Leadership Board
Ranked leaderboard tracking all contributors — Selector, Annotator, Reviewer points
Validation
Usability Testing
Due to the constraints posed by the COVID-19 pandemic, we conducted online surveys and telephone interviews to gather valuable insights. Using this data, I created low-fidelity mockups through paper sketching, which were then refined into high-fidelity mockups. For usability testing, we utilized the System Usability Scale (SUS) to evaluate and measure the user experience.
The System Usability Scale (SUS)
Participants were asked to score the following 10 items with one of five responses ranging from Strongly Agree to Strongly Disagree:
1
I think that I would like to use this system frequently.
2
I found the system unnecessarily complex.
3
I thought the system was easy to use.
4
I think that I would need the support of a technical person to be able to use this system.
5
I found the various functions in this system were well integrated.
6
I thought there was too much inconsistency in this system.
7
I would imagine that most people would learn to use this system very quickly.
8
I found the system very cumbersome to use.
9
I felt very confident using the system.
10
I needed to learn a lot of things before I could get going with this system.
System Usability Scale
85.5
/ 100
V1 Research Baseline · Above Average for First Build
Industry Benchmark Scale
Unacceptable
Poor
Good
Excellent
0
50
100
85.5
Discovery Clue V1
68
Average SUS (all products)
58
Typical annotation tool
67.6 exceeds the average annotation tool benchmark and aligns with the global SUS average — a strong foundation for the next iteration.
Outcome
Impact & Reflection
The result was an application that not only improved the accuracy and efficiency of the annotation process but also encouraged data scientists to engage with it in a more meaningful way. Through a combination of clear design, user-friendly features, and continuous user feedback, we were able to build a tool that enhanced data quality and ultimately contributed to better deep learning model performance.
The application became a key part of the data scientists' workflow, improving their overall satisfaction and productivity. The gamification elements proved particularly effective — the Leadership Board and reward system created a sense of community and achievement that sustained engagement across long annotation sprints.
👥
3
User Roles Defined
📊
85.5%
SUS Score (V1)
🚀
↑
Team Productivity
UX Case Study
Discovery
Clue
Redesigning data annotation for deep learning — turning a tedious, error-prone workflow into an engaging, gamified mission platform.
UX / UI Design
User Research
Prototyping & Testing
Aug 2020 – Jun 2021
3
User Roles
67.6
SUS Score
1 Year
Duration
Sign In
Enter your credentials to access your workspace.
Login
Sign Up
Email / Username
researcher@institution.edu
Password
••••••••••••
Keep me signed in
Forgot password?
Sign In
Or continue with
Microsoft SSO
Google Workspace
9:41
Mon, Apr 27
79%
Welcome back,
Researcher
Your annotations help accelerate drug target discovery for millions of patients worldwide.
"Discovery Clue changed how our team manages annotation workflows. The quality of data is remarkable."
— Dr. S. Müller, ETH Zürich
9:41
Mon, Apr 27
79%
Discovery Clue
Research Annotation Platform
Disease
Target
3
Role
4
Paper
Choose your difficulty level
Cystic Fibrosis → CFTR Trafficking · Step 3 of 4
Selector
Beginner
+50 pts / paper
Review and categorise research papers by Disease, Gene, and Variant type before they enter the annotation queue.
Key Skills
Paper classification
PubMed navigation
Domain literacy
SELECTED
Annotator
Intermediate
+100 pts / paper
Label Disease–Gene–Variant–Assay–Effect relationships within selected paper abstracts and full text.
Key Skills
Variant interpretation
Gene ontology
Structured data entry
Reviewer
Expert
+200 pts / review
Validate and approve annotations from Selectors and Annotators. Final quality gate before data enters the pipeline.
Key Skills
Clinical expertise
Variant curation
Data QC & audit
Back
Continue
9:41
Mon, Apr 27
79%
Discovery Clue
·
Research Platform
847 pts
Good morning
Amir Karimi 👋
Lv 3 · 1,840 / 2,700 XP
3
Missions
#12
Rank
24
Done
1,840
Active Missions
View all →
Cystic Fibrosis
CFTR Trafficking · 3 / 7 steps
+500 pts
Alzheimer's Disease
Tau Propagation · 1 / 5 steps
+300 pts
Breast Cancer
BRCA1 Expression · 2 / 6 steps
+400 pts
Available to Join
2 NEW
Browse all →
Parkinson's Disease
NEW
LRRK2 Pathway · 4 papers · BioRxiv
+300 pts
+ Join
Inflammatory Bowel
NEW
TNF-α Signalling · 4 papers · IJR
€10 voucher
+ Join
9:41
Mon, Apr 27
79%
Disease
2
Target
3
Role
4
Paper
Select your Disease Mystery Game
Your annotation effort directly accelerates drug target identification. Choose the disease area you'd like to support.
Cystic Fibrosis
CF · CFTR · Fibrosis
142 papers
38 genes
Inflammatory & Metabolic Disorders
IMD · Metabolism
89 papers
21 genes
Alzheimer's Disease
AD · Neurodegeneration
214 papers
63 genes
Breast Cancer
BC · Oncology · BRCA
178 papers
54 genes
Type 2 Diabetes
T2D · Metabolic
103 papers
29 genes
Parkinson's Disease
PD · Neurodegeneration
96 papers
31 genes
Back
Continue
9:41
Mon, Apr 27
79%
Discovery Clue
Annotate.
Accelerate.
Discover.
Label Disease–Gene–Variant relationships in medical papers and fuel the next breakthrough in precision medicine.
Get Started
Already a member? Sign In
12K+
Papers
98%
Accuracy
420
Researchers
Architecture
Information Architecture
A structured map of all six platform areas — showing how the app flows from the Homepage hub through each section's screens and role-based sub-states.
Discovery Clue — Navigation Map
Entry / Hub
App Area
Screen
Sub-state
Entry
App Area
Screens
Sub-state
🏠
Homepage
Authentication
Dashboard
Mission Setup
User Roles
Annotation Workspace
Rewards
Login / Sign Up
Onboarding
Active Missions
My Stats & XP
Leadership Preview
New Missions
Disease Selection
Mission Target
Selector
Annotator
Reviewer
Protein Q&A
Full Paper Reading
Change Paper
Receiving XP
Leadership Board
Answers Questions
© 2026. All rights reserved.
Muhammad Inzimam Saghir