← All work

End-to-end UX · Academic Team Project

SnapBudget

Reducing expense-logging friction through a scan-first flow that automates the tedious parts without taking correction and control away from the user.

Context
12-week UX studio project for Canadian university students
Team
Five-person team · I served as team lead
My role
UX research, end-to-end interaction logic, prototyping, and entry-flow design
Result
High-fidelity Figma prototype · Instructor score 93/100

01 · Problem framing

Budgeting was a friction problem.

For university students managing fluctuating income, textbooks, daily purchases, and tight schedules, the perceived benefit of logging an expense often did not justify the effort of entering amount, date, merchant, and category.

Reduce effort first. Budgeting follows.

I helped plan the research, synthesize the evidence, and coordinate the team around one core task. Ten interviews showed three recurring patterns: manual entry felt like homework, students wanted tools that required less ongoing attention, and tracking often stopped during busy weeks.

Discovery evidence · 10 semi-structured interviewsThree patterns narrowed the project to the repeated action that caused students to disengage.
Interview notes, team brainstorming, sketches, and synthesized findings

Interview notes, affinity synthesis, and early sketches

Student expense logging journey map

Journey map locating the point where effort interrupted the habit

02 · Research to decision

Turn evidence into a design principle.

Research indicated that manual entry felt disproportionate to the value students received. I sketched the first complete low-fidelity flow, used team critique to refine its logic, and helped establish receipt scanning as the shared core concept.

I proposed voice input after reviewing accessibility guidance. It became an additional path for users who may find photo or manual entry difficult, while manual correction remained available when automation was inaccurate.

  • 01Prioritize capture speed over dashboard complexity.
  • 02Use the camera or voice to bypass keyboard-heavy entry.
  • 03Keep automation visible so users can verify and correct what the system inferred.
Evidence → decisionThe research changed the interaction model, not only the interface style.
FINDING

Manual entry felt heavier than the purchase.

PRINCIPLE

Reduce effort before adding budgeting features.

DESIGN MOVE

Capture first; keep inference reviewable.

03 · Core interaction

Capture without a form.

The high-fidelity prototype replaces one long expense form with three entry choices: photograph a receipt, describe the expense by voice, or edit it manually. Photo and voice reduce typing; manual entry remains available when automation is not the right fit.

The key product decision is the review step. SnapBudget shows what it inferred—merchant, category, amount, and date—before anything changes the budget, so speed never removes user control.

01 · CHOOSE

Match the moment

Photo, voice, and manual entry are presented as equal starting points.

02 · CAPTURE

Reduce typing

Receipt guidance or a spoken sentence gathers the needed information.

03 · REVIEW

Make inference visible

Parsed details stay editable before the user explicitly confirms them.

04 · FEEDBACK

Close the loop

Updated expenses and savings make the result of the action immediately visible.

Prototype: Final · High-fidelity Figma framesOne short story: overview → input choice → review → updated balance.
SnapBudget overview showing expenses, savings goal, and history01

Overview

SnapBudget menu offering photo, audio, and manual expense entry02

Choose input

SnapBudget review screen for checking inferred expense details03

Review & confirm

SnapBudget updated balances and monthly expense feedback04

See the result

Interactive prototype walkthrough · 00:40Watch both low-effort routes and the explicit confirmation loop in context.
  1. 01Dashboard
  2. 02Photo or voice
  3. 03Review & confirm
  4. 04Updated balance

04 · Validation

Low effort still needs high confidence.

Ten moderated think-aloud sessions shifted attention from speed alone to comprehension. Participants liked having photo, voice, and manual entry, but several were unsure what each option did and whether typing would still be required.

  • 01Clarify the primary actions on the home screen.
  • 02Show an example before users begin speaking.
  • 03Replace ambiguous symbols with explicit confirmation language.
10 moderated think-aloud sessionsEach comparison isolates the interface change; the finding remains readable underneath.
Before: dashboard with decorative controls competing for attentionAfter: dashboard focused on photo and audio expense entry
01 · HIERARCHY

Make the real actions dominant.

Participants confused decorative elements with controls. The revised home screen removes false affordances and centers photo and voice entry.

Before: voice-entry concept with a receipt-style transcriptionAfter: voice entry with a concrete example
02 · EXPECTATION

Show what useful input sounds like.

A concrete spoken example reduces hesitation and explains how the app will turn a sentence into expense details.

Before: ambiguous check and cross confirmation controlsAfter: explicit yes and no confirmation controls
03 · CONFIDENCE

Replace symbols with explicit choices.

Clear Yes and No language replaces icons whose meaning participants had to infer.

Outcome and reflection

The project earned 93/100 for analytical rigor, clear communication, and evidence-led design.

Leadership

I coordinated a five-person team while contributing across research, synthesis, flow design, prototyping, testing, and iteration.

Research synthesis

I connected ten interviews to three recurring patterns and one focused product principle.

Accessibility initiative

I proposed voice input as an additional route after identifying an accessibility gap in photo and manual entry.

Usability-led refinement

I used ten think-aloud sessions to improve expectations, feedback, and confidence across the core task.

TRADE-OFFS AND NEXT STEP

Automation reduces effort, but receipt and voice recognition can be inaccurate, financial data is sensitive, and speech is not suitable in every context. A next iteration would test correction effort, privacy expectations, and the manual fallback earlier with rough prototypes.