Overview

The loneliness problem no app had solved for international students

International students consistently face an invisible wall when adjusting to a new campus. Transportation fades as a challenge over time, but social isolation doesn't. No existing platform combined trusted event discovery, peer assistance, and community discussion in a way designed specifically for this group.

We set out to build something that felt safe, motivating, and genuinely useful from day one on campus.

Research

20 students. One consistent pain point.

We ran surveys with 11 students and in-depth interviews with 9, focusing on two primary problem areas: socialization and transportation. The data quickly told us where to focus.

• Students struggled most to connect outside their immediate cultural group, as language barriers made casual socialising feel risky.
• Existing apps lacked trust signals. Random meetups felt unsafe, and no platform was built specifically for this audience.
• Motivation to keep using a social app dropped sharply without a reward loop or reason to return daily.

Design Decisions

Four features, each solving a specific research finding

High-Fidelity Designs

Onboarding & personalised feeds

The onboarding flow collects interest signals upfront to personalise the event feed immediately, removing the empty-state problem that causes drop-off in social apps.

Host & join events

Event cards surface key details at a glance: date, location, and who else is going — verification badges signal trust without extra friction.

Spaces & rewards

Topic-based spaces organised by category, and a rewards dashboard that makes progress visible, both designed to keep students coming back.

Usability Testing

Testing confirmed the core concept and surfaced real fixes

Three participants each completed five tasks: creating a profile, hosting an event, browsing Spaces, exploring their profile, and redeeming a reward. Sessions ran 45 minutes each.

• The rewards and events features were the most positively received, as users called them intuitive and motivating.
• "Spaces" needed a clearer label; the term confused participants unfamiliar with the concept.
• The "Count on me" button wording caused confusion and was flagged for a language revision.

Outcomes

Deliverables

Overview

A website that frustrated the very students it existed to help

Collegescholarships.org serves millions of students seeking financial aid, but its cluttered interface, poor information hierarchy, and absent navigation made the core job nearly impossible. Content overwhelmed rather than guided. Important actions had no visual emphasis. Users gave up before finding what they came for.

I led this redesign from heuristic analysis through to high-fidelity, guided by a single principle driving every decision: simplicity in service of the user's actual goal.

Research

Every single participant struggled before I changed a pixel

I ran moderated usability testing with 5 participants across three user groups — remote and in-person — before designing anything. 8 tasks were tested covering navigation, search, eligibility checks, and application. Every design decision that followed was grounded in what I observed, not assumption.

• Navigation was the highest-severity issue (score 60). Every participant struggled to find relevant information — the top nav was described as unhelpful, and there was no way to recover when users got lost.
• Search and filter was the lowest-performing task with only a 60% success rate. Participants ran into errors they couldn't recover from without assistance.
• Content volume was overwhelming across all 5 participants. Think-aloud sessions captured this directly: "I don't wanna go through these 300 pages" and "This information is too much to process."
• 100% of participants were first-time users and 100% left dissatisfied — yet 80% said they would still recommend the site to a friend, confirming the need was real even if the experience wasn't.

User Groups

Three distinct users all underserved by the same broken experience

Low-Fidelity Wireframes

Restructuring the information architecture first

Before any visual design, I mapped a new IA that guided users directly to their goal — scholarships, grants, or loans without nested navigation or dead ends.

High-Fidelity Designs

Before → After: Navigation

The original site was an unending nest of information with no structure. I reorganised major sections to create a direct path to scholarships, loans, and grants, with a resource hub that was always one click away.

Before → After: Search & filtering

The original search returned an unending scroll with no way to narrow results. I introduced intuitive sort and filter controls, popular search suggestions, and paginated results with clear user control at every step.

Scholarship detail page

Each listing now surfaces eligibility criteria, deadline, award amount, and application information in a structured layout so students can qualify themselves at a glance before investing time in an application.

Outcomes

Deliverables

Overview

Essentials delivery on demand, even at odd hours

OjaNow was founded by three Nigerian entrepreneurs who identified a clear gap in the market: no reliable, prompt service existed for delivering groceries and daily essentials when people actually needed them. The platform stocks over 300 products across strategic locations to keep delivery times within minutes, not hours. Designing for the Nigerian context added a specific layer of complexity — inconsistent street addressing and the near-absence of reliable mapping data make rider-to-customer navigation a real operational challenge, not just a UX one.

I led design across the entire mobile experience: the customer app and the rider app, while overseeing the design direction for the web and admin interfaces.

The Challenge

Building for conversion and retention from day one

The core business challenge wasn't just building an app but building one that converted visitors into buyers and kept them coming back. I joined after the initial stakeholder alignment and market research, reviewed the product documentation and existing findings, then collaborated with the product manager and technical lead to define the design strategy.

Conversion rate and retention were the two metrics the business cared about most. I kept both front of mind through every design decision, from the browse experience to the checkout flow to the rider handoff.

Business Goals

Targets set before launch

User Flows

Mapping both sides of the transaction

I designed separate flows for the customer and the rider, each with distinct goals, contexts, and success states, then ensured the handoff between the two was seamless. The customer flow covers discovery through to delivery confirmation; the rider flow handles order acceptance through to completion.

Customer App

Browse, order, track in as few taps as possible

The customer experience is built around speed and clarity. The home screen surfaces categories and promotions immediately. Product pages are stripped of friction. Checkout is optimised for the ₦1,000–5,000 basket size the business was targeting for quick, low-commitment purchases that don't require deliberation.

Real-time order tracking with rider details keeps the customer informed and reduces anxiety between checkout and delivery, one of the biggest drop-off triggers in delivery apps.

Order tracking

The tracking screen gives customers a live map view, estimated arrival time, and direct access to the rider — reducing support load and building confidence in the product promise of delivery in minutes.

Rider App

Designed for speed, clarity, and zero confusion mid-delivery

The rider app operates in a completely different context where the rider is in motion, often in high-pressure situations. The interface prioritises large tap targets, minimal reading, and clear status progression. Riders receive orders, navigate to the customer, and confirm delivery without switching between apps or making phone calls.

A key constraint specific to the Nigerian market: street addressing is inconsistent and mapping coverage is unreliable in many areas, making turn-by-turn navigation alone insufficient. The design accounts for this by surfacing the customer's landmark descriptions and a direct call shortcut prominently — so riders can resolve location ambiguity quickly without breaking their flow.

Product Evaluation

10 participants. Strong signals before launch.

Usability testing was conducted on a prototype with 10 participants. Beyond the headline numbers, watching users navigate the app revealed specific friction points, areas where hesitation or confusion showed up that the metrics alone wouldn't have caught.

• The remaining 10% revealed friction points in the flow that were addressed before launch; the testing paid for itself immediately.
• The NPS indicated strong likelihood to recommend the app, validating the core value proposition even at prototype stage.

After Launch

The product exceeded every business target

Post-launch data showed the design decisions translated directly into business performance. The conversion rate beat the original target by over 2× and active users grew to 500+ — validating both the product concept and the design approach that prioritised frictionless purchase behaviour.

Deliverables

Overview

Design as advocacy, targeting those with the power to change the system

Police brutality in Nigeria is a longstanding human rights crisis of extrajudicial killings, unlawful arrest and detainment, physical assault, extortion, and rape have fuelled sustained public outcry from citizens and international communities alike. The disbandment of SARS in 2020 following nationwide protests showed that pressure works. But systemic change requires sustained, targeted communication.

This campaign was designed not to preach to those already aware, but to reach policy makers and decision makers, the people with the power to reform the police and build structures that make brutality practically impossible.

Research

10 participants. Their experiences shaped every design decision.

I conducted qualitative research with 10 participants who shared firsthand accounts of police brutality. Their experiences informed both the tone and the strategic direction of the campaign.

• Participants converged on three solutions: accountability structures, police re-education, and adequate funding, all requiring policy-level action.
• This validated the campaign's strategic target: not public awareness alone, but decision makers who control reform levers.

Problem Mapping

A tree-structure breakdown of the problem space

Using the research findings, I mapped a tree-structure diagram breaking the problem down into its root causes, contributing factors, and potential solutions, giving the campaign a clear strategic foundation before any design work began.

Design Direction

Bold, urgent, impossible to ignore

The visual language of the campaign was deliberately confrontational with high contrast, bold typography, and a palette rooted in the Nigerian flag's green and white alongside urgent red and yellow. The goal was material that demanded attention in both digital and physical contexts, not content that could be scrolled past.

Outputs spanned three channels: a campaign website, digital social assets, and print materials, each adapted for its medium while maintaining a unified visual identity.

Website Mockups

A reporting platform: "Don't be a passer-by"

The campaign website gave citizens a direct action: report an incident. The headline "Don't be a passer-by — report what you witnessed" framed civic responsibility as the call to action, lowering the barrier between witnessing brutality and doing something about it.

Print & Digital Artifacts

Research reports, manifestos, and shareable assets

Print materials were designed to work as standalone advocacy pieces, including a research report booklet covering findings and proposed solutions, and a manifesto piece with the message "A Better Police Force. A Better Society." Each piece was formatted for both digital distribution and physical printing.

Digital campaign assets

Social and digital formats carried the same bold visual language with high-contrast compositions, urgent typography, and a direct call to report or learn more.

What Didn't Work

The body tag idea and why user feedback killed it

The original concept included a wearable body tag with a QR code that people could scan to report incidents or learn more. In theory it was compelling as activism you could wear, but in practice, user feedback revealed a real problem: many people felt uncomfortable with the idea of scanning a barcode placed on someone's body.

Rather than push the concept through, I pivoted to stickers with the same QR code functionality, applied to any surface the person chose. Same reach, zero discomfort. This is the kind of decision that separates research-driven design from design that imposes an idea on users.

Deliverables

Overview

The misinformation problem is a design problem

The speed and scale at which information travels through social media has created a landscape where false information spreads faster than corrections. Misinformation affects politics, health, religion, and personal wellbeing — yet solutions that rely purely on AI moderation risk suppressing legitimate speech.

This project tackled a specific design question: how do you build a pre-dissemination AI verification system that users actually trust and accept — one that checks content before it spreads, not after?

Problem Framing

Three constraints that shaped every decision

Stakeholder Analysis

Mapping who holds power — and who bears the risk

Using a stakeholder register and power/interest map, I identified eight distinct groups with competing goals: from social media platforms and government bodies with high power, to users and ethical advocates with high interest but lower direct influence. This mapping shaped which voices were prioritised in the co-design process.

Research

Three themes from literature — validated through expert interviews

Literature review using both deductive and inductive coding identified three recurring themes. These were stress-tested through an in-depth interview with an AI researcher and release train engineer specialising in misinformation detection.

Persona & Journey Mapping

Jessica, a daily social media user trying to stay informed

I developed a persona: Jessica, 29, an industrial designer, who uses social media daily for professional and personal news. She's familiar enough with AI to appreciate its potential, but wary of its decision-making when it touches on human values. Her journey map traced the emotional arc from entering the platform to encountering a suspicious post and realising she couldn't verify it, the moment of peak frustration that the intervention targets.

The Solution

Pre-dissemination AI verification as a spell-check for truth

The core concept: before a user posts, the AI analyses the content for potential misinformation. If flagged, the user sees conflicting sources from the web, not a block but a prompt. This preserves the user's choice while giving them the information to make it responsibly. The metaphor driving the design was "AI as spell-check" that flags, informs, and never overrides.

• Fact-checking and verification: AI combs datasets to assess whether a post is false or misleading, flagging for human authentication if needed.
• Crowdsource verification: users can flag content themselves, routing it to AI or human moderators for review.
• Counter-narratives: alternative information sources surface alongside flagged content, providing context rather than censorship.
• Human-in-the-loop: AI and human moderators form a feedback loop, with humans handling edge cases and training the model continuously.

Simulated Prototype

A simulated X platform to test the AI verification flow

Due to technical constraints, a fully functional integration of the Meta Sphere AI tool wasn't completed. Instead, I built a simulated X platform that replicated the verification flow, showing users exactly how the AI interaction would behave before posting. Participants walked through the flow, triggering the AI check, seeing flagged content, and navigating conflicting source prompts.

Triptych Emerging Solution

Opportunities and to-dos from the journey map

The journey map surfaced a clear opportunity: users had no way of knowing whether content they were reading was skewed or misleading. I extracted a clear "To Do" from this: design a warning label that surfaces when a post may be misleading, and test whether users find it helpful rather than intrusive.

Data & Model Validation

Training a misinformation classifier to validate the concept

To ground the design in technical reality, I trained and evaluated machine learning classification models on a 2,045-article dataset sourced from Kaggle. The goal was to understand what AI-driven misinformation detection is actually capable of, and where it falls short, so the UX could be designed around those real constraints, not an idealised version of the technology.

• The Random Forest model (78.7%) outperformed Logistic Regression (72.1%) in both accuracy and F1-score across fake and real classes.
• The class imbalance (63.1% fake vs 36.9% real) created recall challenges for the "real news" class, a critical real-world concern that informed the human-in-the-loop design requirement.
• These limitations directly validated the design decision to make AI a prompt, not a gatekeeper. A 78% accurate model cannot be trusted to make final calls alone.

User Testing

5 participants. Generally positive results, though trust was earned, not assumed.

Five participants engaged with the simulated app, completed a post-engagement survey, and participated in follow-up interviews. The findings were nuanced: users understood and appreciated the AI verification concept, but trust varied based on concerns about accuracy and bias.

• All participants had moderate trust in the AI system, and no one fully delegated decision-making to it, which validated the design's non-blocking approach.
• The most common concern was AI accuracy: users wanted to know the system's error rate before trusting it on sensitive content.
• Privacy concerns were mixed; some participants were unconcerned while others wanted explicit data usage policies surfaced in the UI.

Outcomes

Deliverables

Real-World Validation

X launched exactly this and called it Grok

In November 2023, xAI (Elon Musk's AI company) launched Grok, an AI assistant embedded directly into X (formerly Twitter). Its core use case? Real-time information verification, surfaced at the point of consumption. This project proposed an AI that checks content before it spreads and surfaces conflicting sources for the user to decide, which became one of the most-used features on one of the world's largest social platforms.

Grok's integration into X gives it something no other AI chatbot has: direct, real-time access to the platform where misinformation moves fastest. Users invoke it mid-scroll to verify claims, interrogate news posts, and stress-test information before sharing it, exactly the human-in-the-loop pre-dissemination verification pattern this project advocated for.

Why Grok's success directly validates this project's thesis

• Embedded, not external: Grok lives inside X, exactly where misinformation originates. This project argued the intervention had to be at the point of dissemination — not a separate fact-checking website users would never open.
• Human-in-the-loop by design: Grok surfaces information and lets the user decide. It doesn't block posts or override judgement. This project's core principle, AI as prompt not gatekeeper, is reflected in every Grok interaction.
• Real-time access is the differentiator: Grok's unique advantage is live X data. This project identified real-time detection as the critical technical requirement; xAI built their entire infrastructure around exactly that.
• Trust through transparency: Users spend nearly 14 minutes per session with Grok. This project's user testing found transparency about AI's role was the single biggest driver of user comfort. Grok's growth confirms it.

Users on X, invoking Grok to verify in real time

Screenshots below show real X users using Grok to fact-check posts, interrogate claims, and surface conflicting sources, the exact user behaviour this project set out to enable.