Impact
+127% increase in user engagement
Post-launch, we saw an impressive 147% increase in feature engagement, signaling strong product market fit in the overall user experience
40% of Users Reported Less Stress
Users told us they were feeling much more confident and in control over their wedding budget and estimations after launch
Context
Wedding planning consistently exceeds budget expectations. Nupt.AI uses AI to streamline the process, but the budget estimation tool became a critical drop-off point. Users abandoned the tool because they couldn't trust the numbers and felt constrained by rigid categories that didn't reflect their reality.
As a product design intern and sole designer working on this feature, I handled all aspects of the design process from initial research to design, prototyping and testing.
Additionally, I created over 15+ reusable components and icons that could be used in the future.
Wedding budgeting is stressful and the original tool at Nupt.AI was making it worse
Couples were dropping off before they even finished the estimator. The interface was a wall of static numbers with no flexibility, no context, and no forgiveness. If you changed one figure, the rest of the math broke. If you didn't know what was "normal" to spend on a venue in your city, you were on your own.
User research surfaced three clear pain points:
No flexibility — users couldn't estimate at a high level before drilling into line items
Fear of being wrong — without location benchmarks, every number felt like a guess
Broken logic — edits didn't cascade intelligently, leaving the budget in an inconsistent state
Primary and Secondary Research
Before touching any designs, I grounded myself in the problem through a mix of primary and secondary research. On the primary side, I ran usability tests with 5 users to evaluate early "lock" and "toggle" concepts, gathering clarity ratings and qualitative feedback that directly shaped our design pivots — with iterative testing continuing throughout each phase. I also ran A/B tests to validate decisions by measuring engagement and preference between feature variations.
For secondary research, I analyzed The Knot and Zola to map existing mental models, comparing their feature sets, visual approaches, and user flows. This surfaced what each did well — The Knot's location-based data intelligence and Zola's visual spend tracking — and where both fell short. I paired this with a heuristic evaluation of the existing budget tool, identifying clear violations in flexibility, feedback, and error prevention, alongside market research into wedding planning trends and Gen Z design preferences that later informed the visual direction.
Three core friction points emerged from this work: users wanted to estimate at a high level before committing to line items, they feared entering numbers without knowing what was "correct" for their area, and the tool's rigid logic meant a single edit could break the entire budget. The impact was measurable — 48% of couples reported delays and a lack of feedback, eroding trust in the tool before they'd even finished using it.
What I Built
I led the redesign from concept to developer handoff, working in weekly syncs with our founder, design and engineering team to make sure every design decision was both feasible and purposeful.
The final solution centered on three things:
Smart AI Estimation — Instead of asking couples to guess, the tool pre-fills budget categories based on their total budget and location. No blank canvas, no paralysis.
Two-tab structure (Estimate → Payments) — Couples could play freely with numbers in "Estimate" before committing anything in "Payments." This separation was key to reducing anxiety.
Intuitive status indicators — Rather than complex locking or toggling mechanics (both of which tested poorly with users — 40% rated the lock feature 2/5 on clarity), I replaced them with simple Over/Under Budget indicators. The AI handles the math. Users just edit freely.
Decisions Worth Calling Out
Killing the lock feature. Early designs included a lock icon to let users pin certain categories while AI recalculated the rest. It seemed clever. Users hated it. Surveying just 5 participants made it obvious — both the lock and toggle concepts added cognitive load without solving anything. We cut it and never looked back.
Choosing the right model. Picking an LLM for the estimator wasn't just a tech decision — it was a product decision. Gemini 2.5 Flash Lite was fast but broke budget constraints. ChatGPT 5.2 was inconsistent. We landed on Gemini 2.5 Flash: slower (30–45s), but 95% accurate. Trustworthy numbers were worth the wait.
Injecting personality into the UI. Late in the process, I noticed the category cards felt utilitarian — functional, but joyless. Wedding planning should feel exciting. I drew inspiration from beauty and wellness brands and introduced a warmer palette, rounded arch motifs (a nod to wedding arbors), and color-coded categories. The visual upgrade wasn't cosmetic — it shifted how users felt about the tool.
Working Well Under Pressure
For this specific feature cycle, I had a short window of time to research, design and ship, so I needed to think outside the box. Weekly syncs with stakeholder teams helped keep everyone aligned on the end goals.
I knew from listening to our users that the current design felt "static" and "off", so it was crucial for me to bring clarity to the interface and ensure our users could trust our estimations. In addition to the design, I partnered closely with our engineer to ensure we selected the best LLM that could generate accurate estimations, which was Gemini Flash 2.5. It was important for us to prioritize accuracy over speed.
I started by conducting primary research by assessing current market trends, identifying UX patterns across key players like The Knot, Zola, and other household names in the wedding planning space. I also expanded my thinking to include beauty and wellness platforms like Ulta, Sephora, etc. to help hone in on our target user.
After multiple iterations, feedback from stakeholders and the team at FoundersBoost, I landed on my final proposed design. I sought to soften the overall features on the page, not just by "rounding corners" but really taking into consideration the overall experience for our users, making sure the interface felt friendly, approachable, engaging but not overwhelming. I carefully selected colors within our existing design system to incorporate eon to the Budget page, so that we maintained consistency across the platform.
Solution
I led the redesign through weekly syncs with engineering and development teams, evolving from a static form to a dynamic two-tab system: "Estimate" and "Payments."
Smart Estimation
AI pre-fills categories based on budget and location, eliminating blank canvas paralysis while giving users a realistic starting point.
Technical decisions
We tested three AI models for budget calculation. Gemini 2.5 Flash Lite was fast (10-15 seconds) but failed budget constraints. ChatGPT 5.2 offered reasonable performance (20-25 seconds) but struggled with instruction adherence. We selected Gemini 2.5 Flash—slower at 30-45 seconds but achieving 95% accuracy within constraints. We prioritized trust over speed.
Design evolution
Early iterations tested "lock" and "toggle" features to fix budget categories. User testing with 5 participants revealed both concepts confused users (40% rated clarity as low). We eliminated the feature entirely, pivoting to let AI handle calculations while users edit freely.
Visual design drew inspiration from beauty and wellness industries to elevate the UI from calculator to experience. I introduced rounded arch motifs (nodding to wedding arbors) and a warm, sophisticated palette that resonated with our Gen Z audience.
Demo Day @ Evite HQ
With Nupt.AI being part of an 8-week accelerated program through FoundersBoost Glendale, this gave our team the opportunity to showcase our product and features in front of investors, founders, mentors and other entrepreneurs working in the tech industry. I had an amazing time speaking with other startups and founders, and look forward to many more experiences like this in the future.
