Nova
ROLE
Founding Product Design Lead
TEAM
Lauryn Kinsella
Devin Hayden
Teri Shim
TIMELINE
Oct 2024 - Feb 2025
TOOLS
Figma
OVERVIEW
No-internet schools lack resources for quality education.
A handful of underfunded schools in South Sudan and Tanzania, educators complained about unreliable internet, insufficient materials, and difficulty keeping up with changing curriculums. These problems had prevented students from getting into secondary school and pursuing higher education.
APPROACH
Nova: offline learning platform
A device that broadcasts an intranet (think WIFI), giving nearby devices access to an edge computed AI tutor and learning resources including Khan Academy and Ted Talks.
OUTCOMES
Shipped to Tanzania and South Sudan
Collecting quantitative data in South Sudan
100+ learning resources
Open resource offline learning
50+ devices
First locally-run AI tutor
Trained on their curriculum remotely update-able
This case study focuses on design for students and teachers with low digital literacy while under high hardware constraints.
INITIAL FINDINGS
Limited technology + no internet = difficult independent learning.
"Computer labs are there, but there is no internet and no computers, so it just sat there and stopped me."
Said a high school senior
A Computer Lab with No Computers, No Internet
< 10% of students go to college
The exam-based structure is cut-throat.
Insufficient teachers
Teachers are often underqualified and schools areunderstaffed.
Outdated material
The government keeps changing curriculums, and teachers have a hard time keeping up.
Unreliable/No Internet
Schools are underfunded and the government is concerned with internal conflict = No Starlink.
WHAT EDUCATORS SAY
Teachers Need More Resources
Our non-profit partner LetAllGirls talked with an educator in Tanzania to learn about their current setup.
01.
Teachers lack material to accommodate all students.
Director + teacher
"With 50 students in a class, it’s hard to satisfy with the material we have."
Not enough material to go around and limited technology to make up for it.
02.
Difficulty making material for new government curriculum.
Director + teacher
"It would be great if we could make new quizzes based on the curriculum."
3+ schools were understaffed and had to create new material manually with outdated tools.
How do we bring quality, up-to-date resources to students + teachers and encourage curious exploration?
Curriculum-trained AI tutor
The tutor was to be dynamic with not only the curriculum but to each student’s education level.
Quiz Generator
Help teachers focus on students rather than on constantly reviewing large syllabus updates.
Phet Simulation Lab
Schools often couldn't afford lab equipment, but teachers wanted to promote group work and hands-on activities.
Searchable Digital Library
As opposed to the piles of hyperlinks provided by the RACHEL (only other offline database), our library had to be easy to navigate.
Smart Companion
Ask about what’s on screen. No hopping tabs.
ADDRESSING THE GAP
Existing solutions didn't fuel curiosity
Students hardly benefitted from existing offline databases due resources being dated and difficult to navigate.
Evaluating existing solutions
(NOVA concept 3D render). Easily update-able, low maintenance, and explorative at a cultural level.
Competitors are Outdated or Internet-reliant
ONLINE SOLUTIONS
Assumes access everywhere.
Relied on the fact that students had internet and devices at home, but it's not the case.
Unable to implement offline.
Difficult to economically produce at scale.
OFFLINE SOLUTIONS
Outdated material, crappy navigation.
Existing hardware only supported disorganized hyperlinks to content that wasn't culture-relevant.
EXISTING WORKAROUND
Students are curious to do their own research independently.
Students borrow teachers’ phones just to Google search, despite poor connectivity.
ux under technical pressure
Navigating high technical constraints posed by edge computing.
(extremely low-cost, custom-built offline computer).
Ideal Platform vs. Necessary Features
Designers went crazy on whiteboards then discussed with engineers to narrow the scope.
Found core features to focus on according to user needs.
Less computing power for affordability = a sharp focus on providing value.
We had to minimize the amount of “beautification” that would cost development time.
A GPU and motherboard in a precarious pose.
Matching problem to approach
We matched each major problem with an approach
Matching the current structure, not disrupting or reinventing.
Teachers were busy updating their material according to new syllabi, so we focused on integration with any teaching style and assignment.
Managing ai behavior
Hardware limited LLM context, so we designed backend for better UX
Avoid hitting "new chat" every 5 prompts
less available tokens
The budget and hardware lowered our LLM's available tokens. At first, we expected users to press “new conversation” every 5 prompts…
Users would be frustrated at its forgetful nature.
Frustrated User Journey
This performance challenge would cost students to halt exploration or getting help on a project. With no room for a full onboarding or an account setup system, each student wouldn’t know the exact function of a simple “new conversation” button.
I sat down with the lead engineer: Could we prevent the LLM from reading everything top-to-bottom to improve latency?
Success User Journey
A dumb relevance-checking AI
We were able to implement an AI relevance-checking system. Funnily, we had a surplus of storage, so we added an additional "dumb" AI to check and return Y/N, using less computing power and speeding up response generation.
applying Cultural Context
Matching existing habits with similar design patterns
I knew that given 1) their overall lower digital literacy and 2) their very different culture, we’d have to be intentional with each design decision. After interviewing teachers and students, we learned that they were familiar with WhatsApp.
We mimicked WhatsApp’s iconography to improve usability and lean on existing intuition.
result
Shipped to Tanzania and South Sudan
Shipped to 2 schools and 1 library for high school students! Unfortunately had to pass off all software to partner without collecting longitudinal data.
Reflection
What I learned
Ask the dumbest questions
Because everyone ends up assuming limitations into existence, hindering creative thinking.
Think with the tech, not around it.
Learning from engineers gives opportunity to design outside of the “technical box”, making limitations less rigid than imagined.
Make, take, and present notes.
Documenting knowledge on all fronts such as LLM performance and stakeholder goals is key to team alignment and integrated design thinking at every step.
Looking Forward
Personalization
Make student/teacher accounts so the AI tutor and teachers can understand student progress.
Edge-Computed AI
Without the technical constraints, AI that is computed in-house might be the next level of personalization and private security.