Thảo, UX/UI designer. I untangle complex things until they feel clear enough to act on.
About
nice to meet you
thao.jpgthat’s me
I’m Dương Nguyễn Phương Thảo
UX/UI designer, based in Ho Chi Minh City
I came to design through Management Information Systems. A web design course during my degree was the first time I opened Figma, and I haven’t stopped since. Because of that background, I usually map the process and the data behind a product before I design its screens.
Experience
Nov 2025 – Present
UX/UI Design Intern
Tiger Tribe, a Heineken company
Education
Class of 2026
Management Information Systems
University of Economics and Law, VNU-HCM
Work
small worlds I’ve helped shape
01/05
Shelf Price Intelligence
Web platform
A retail price intelligence platform that turns competitor shelf prices into decisions.
View case study02/05
Draught Analytics
Desktop dashboard
A desktop analytics dashboard for understanding draught performance across many outlets.
View case study03/05
Water Treatment Plant Inspection
iPad + Web
A paperless inspection and approval flow for technicians and plant managers.
View case study04/05
Sustainable Returns App
Mobile app
One place to handle returns from several marketplaces, with greener choices built in.
View case study05/05
ClayCo
E-commerce website
A ceramics shop where a personalised request travels from cart to the store’s orders.
View case studyPlay
a small pause, if you need one
A rally in the park
Hit the shuttlecock on your side. Hit it while it is still high for a power shot. Park buddy tires as the rally goes on.
PlayLab Innovation Challenge 2026
Shelf Price Intelligence
A platform that collects competitor shelf prices automatically and turns them into pricing decisions.
Overview
- Role
- Sole UX/UI Designer
- Team
- 7 people
- Timeline
- 2 months
- Result
- 1st place, PlayLab 2026
- Confidentiality
- Names, data and visuals changed
My role
- I owned the product UX/UI, from the design system to every high-fidelity screen.
- I built the three personas and confirmed them with stakeholders.
- I ran UAT with business stakeholders.
- Backend and AI tested scraping for each retailer in parallel.
Problem
In the US, retailers set the shelf price, so the company cannot see what shoppers pay. Specialists visited stores or retailer sites by ZIP code and merged prices in spreadsheets by hand.
Where it broke
Too narrow
A store visit captured two or three of hundreds of SKUs.
Too costly
Twenty people spent a workday a month on it, an estimated $16,000.
Too slow
By the time data was merged, the market had moved.
Reframing the question
How might we give commercial teams automated, store-level visibility into competitive shelf pricing, so they can negotiate, price and respond with confidence?
The brief asked us to automate price capture. Stakeholders showed us someone still had to notice a competitor move and act in time. That is why the product has a dashboard, alerts and AI insight, not just a data pipeline.
Users & evidence
There were no direct user interviews. I built three personas from stakeholder descriptions, and stakeholders confirmed them.
Marcus
Key account manager
Manager role
Needs shelf price proof before retailer meetings.
Rachel
Pricing analyst
Manager role
Needs to benchmark a portfolio and catch competitor moves early.
Daniel
D&T operations
Admin role
Learns the data broke only when someone complains.
Five commercial use cases
Price auditingCompetitive benchmarkingNegotiation supportGeographic monitoringTrend alerts
How I worked
- 1
Understand the brief
Weeks 1 to 2
Continuous Q&A with stakeholders until we agreed on a direction.
- 2
Build the system
From week 3
I skipped wireframes, built tokens, components and chart patterns, then designed in high fidelity.
- 3
Design to development
Feature by feature
Each feature was agreed with the team, then went to frontend.
- 4
Validate
Stakeholder UAT
I tested the design with business stakeholders.
Process trade-off
High fidelity makes ideas harder to drop. Early feature reviews kept that risk small.
Key decisions
01
One product, two roles
- Problem
- Commercial users and data admins need the same data but very different controls.
- Decision
- Marcus and Rachel share a Manager role. Daniel's Admin role adds data configuration, monitoring and user management.
- Trade-off
- Admins also see the full commercial layer, so the two roles only differ in three modules.
- Why it matters
- People making pricing decisions never have to manage a scraping schedule.
02
Compare prices, not just list them
- Problem
- A retailer can carry a hundred SKUs across nine brands. A rep needs to know what to notice first.
- Decision
- Three layers: portfolio by brand, our SKUs against competitors, and a trend by ZIP code. An AI summary explains the movement.
- Why it matters
- The rep's question is how our price compares, so the gap comes first.
03
Design for the failure case
- Problem
- When a scraper returns a product that matches nothing, a silent guess would corrupt the data.
- Decision
- The record is flagged and sent to an admin queue. Admins can accept it, map it to another SKU or create a new one.
- Trade-off
- Unmatched prices wait for a person, so they reach the dashboard later.
- Why it matters
- Bad matches are fixed by a person before they reach any pricing decision.
Final solution
The same data serves two jobs: deciding on prices and keeping the data right.
Spot a competitor move
A manager goes from an alert to the SKU detail behind it.
Fix a bad match
An admin reviews flagged records before they reach the dashboard.
Outcome
First place at the PlayLab Innovation Challenge 2026, later presented to the company's Chief Digital & Technology Officer.
Business case
The business case estimated an automated pipeline across four retailers at about $2,700 a month, against about $16,000 for manual collection.
It would cover the full priority catalogue instead of two or three SKUs per store visit.
These are projections, not measured savings.
Validation
UAT feedback asked for enhancements rather than corrections. It led to five refinements for the next build:
- Auto-scaling chart axes
- Clickable summary cards
- Parallel market and retailer filters
- Store-level detail within a ZIP code
- Alerts grouped by retailer and change type
Reflection
what I learned
I worried the Brand Detail table was too dense. In UAT, business stakeholders who work with pricing data every day did not struggle with the dense table. What looked heavy to me looked normal to them. I learned not to simplify a data-heavy screen just because it looks dense to me.
What I would change
Draw the role line further
Manager and Admin see almost the same thing. The boundary could go beyond module access.
Ship fewer modules
Eight modules suited a competition. A real product needs two jobs: see where we are priced wrong, and act on it.
Ongoing product work
Draught Analytics
A desktop dashboard for comparing draught beer performance across many outlets, not one venue at a time.
Overview
- Role
- UX/UI Designer, under a Lead Designer
- Timeline
- Jul 2026 – Present
- Platform
- Desktop web
- Status
- Ongoing, in revision
- Confidentiality
- Names, data and visuals changed
My role
- I design the desktop screens, extending the mobile app's design system.
- I defined the comparison rules behind the charts.
- The Lead Designer brings the stakeholders' requirements and reviews my work.
Problem
A mobile app already helped a bar owner check their own taps. Two groups fell outside it.
Sales representatives
Need to see hundreds of outlets at once to spot which accounts are losing beer.
Multi-venue owners
Need a view of their network, not one bar at a time.
The design problem
Both are comparison problems, and comparison needs room: bigger charts, several outlets on one axis and tables you can sort.
What the product measures
Yield is the share of beer leaving the keg that actually reaches customers. Too low means beer is wasted, and too high usually means short pours, so neither is good.
Users & evidence
Sales representative
Covers hundreds of outlets
Needs to spot which accounts are losing beer and which are doing well.
Multi-venue owner
Runs several venues
Needs one view of the whole network.
Where the evidence comes from
I had no direct contact with sales reps or bar owners. Their needs reached me through the Lead Designer, from weekly stakeholder meetings and comments on the Figma file.
How I worked
There is no single brief. It started with a list of pages and has evolved through each round of the same loop.
- 1
Take in requirements
From the Lead Designer
The Lead Designer brings requirements from the weekly stakeholder meeting.
- 2
Design options
Against each requirement
I design options directly in high fidelity.
- 3
Review and revise
With the Lead Designer
We review the options together, and I revise.
- 4
Stakeholder comments
On a shared Figma page
Stakeholders comment, and new requirements start the next round.
Why I went straight to high fidelity
The mobile app had already set the tokens, components and visual language. The desktop product had to extend that system, not invent a new one.
Key decisions
01
Compare groups by average per outlet
- Problem
- Put a group of sixteen outlets next to a single bar on total volume, and the group line sits so high that the bar flattens against the bottom.
- Decision
- When groups and single outlets share a chart, it measures average volume per outlet. A single outlet looks the same, since its average is its total.
- Trade-off
- It costs some intuitiveness, because people ask for totals first.
- Why it matters
- It is the only unit where a group and one outlet share a scale and the comparison still means something.
02
Remove the ambiguous comparison
- Problem
- In a twenty day month-to-date view, "previous period" could mean the twenty days before, the same days last month, or all of last month. Each gives a different number, and the answer changed with the period picker mode.
- Decision
- I removed it and kept year over year: the same calendar span, one year earlier.
- Why it matters
- Users no longer have to guess which definition the chart used. One comparison people trust beats two where one is quietly wrong.
03
Keep the benchmark still
- Problem
- Search narrows the page to one outlet or brand, and KPIs, charts, tables and the map all recalculate. If the group average moved too, the benchmark would move with the thing being measured.
- Decision
- The group average became a hard rule. It recomputes for filters and time windows, never for search.
- Why it matters
- A benchmark is only useful if it stays still when the view changes.
Final solution
Four pages, each answering a different question about the same network.
Check the network
A daily view of how every outlet is pouring.
Find where beer is lost
From a drop in yield to the outlets and pours behind it.
Shared rules across all four pages
Two-layer filters
A time window says when, and a repeating pattern, such as weekday evenings, says which parts of it. Filters carry across all four pages.
Asymmetric yield bands
Below 97% is red, 97 to 100% amber, 100 to 105% green and above 105% amber again. The same colours appear wherever a yield number does.
Custom groups
Users save outlets as a group, and every page recalculates against it.
Where it stands
The project is still running. Design is in revision, and stakeholders take a working prototype out for testing.
Agreed but not yet applied
Some decisions are queued, such as renaming the Waste column and making volume units consistent between pages. Stakeholders have not reached those sections yet.
Reflection
what I learned
The hard part turned out to be defining what the numbers should mean when filters, groups and time ranges changed. I learned that analytics design is as much about rules as visuals.
The gap I would close next
I work through a Lead Designer rather than directly with sales reps and bar owners. I can confirm a requirement is met, but not that a screen is obvious to someone opening it for the first time on a Monday morning. That is what the prototype testing is for.
University practicum hosted by FPT Software Japan
Water Treatment Plant Inspection
Designing a paperless inspection and approval flow for water treatment plants: an iPad app for technicians in the field and a web app for plant managers.
Overview
- Role
- UX/UI Designer in a five-person BA team
- Timeline
- May to July 2025
- Platforms
- iPad app and web app
- Status
- Concept, not built or tested
- Confidentiality
- Client anonymised, data illustrative
My role
- I designed the iPad and web interfaces, from wireframes to mockups.
- I wrote seven of the client questions in the requirements log.
- Requirement analysis, process models and the roadmap were team work.
Problem
According to the client, equipment inspections at its plants still ran on paper.
Double entry
Technicians wrote readings by hand, then retyped them into Excel.
Manual approval
Managers signed printed forms, and rejections were given verbally.
Missed readings
Nothing flagged a value beyond its limit until someone read the paper.
Lost records
Past checklists were slow and sometimes impossible to find.
The design problem
Connect field inspection and desk-based approval in one traceable flow, while catching abnormal readings before they disappear into paperwork.
Users & evidence
Inspection technician
iPad, in the field
Records readings on a schedule, often with poor light, time pressure and patchy network.
Plant manager
Web, at the desk
Reviews the day's checklists and approves or rejects them.
Where the evidence comes from
We never observed technicians or managers. What we knew about them came from two rounds of written client Q&A.
How I worked
- 1
Gather requirements
Late May to June
Two rounds of written Q&A with the client. I wrote seven of the questions.
- 2
Map the paper process
Team effort
The team modelled the current process and its gaps before any screen.
- 3
Wireframes, then mockups
Late June to early July
I designed wireframes for each role and turned them into mockups.
- 4
Revise
After mentor review
I revised the mockups after the host company's review.
Key decisions
01
Start inspection by scanning the device
- Problem
- Picking the right form and typing device details by hand is slow and risks logging readings against the wrong device.
- Decision
- Scanning the device's QR code confirms it is assigned to the technician and opens a checklist with device, staff and area already filled in. An unassigned device asks for a rescan, and creating a checklist manually stays available.
- Why it matters
- Fewer fields to type, and the wrong device is caught before any data is entered.
02
Catch abnormal readings inside the form
- Problem
- A reading beyond its limit went unnoticed until someone read the paper.
- Decision
- An out-of-range value turns red with its limit shown underneath, and the disabled submit button says why. Submission waits until the technician confirms an incident report.
- Trade-off
- Technicians take an extra step in the field whenever a reading is out of range.
- Why it matters
- Every abnormal reading becomes a recorded incident before the checklist reaches the manager.
03
Separate a working draft from a submitted record
- Problem
- Technicians need to save partial work, but a manager must approve exactly what was sent.
- Decision
- Save Draft and Save and Send for Approval are separate actions, with a Draft label in the header. A submitted checklist locks and receives an auto-generated ID.
- Trade-off
- Once sent, a checklist can only be corrected if the manager rejects it.
- Why it matters
- What the manager approves cannot change afterwards, and the ID makes the record easy to trace.
Final solution
Checklist status and record IDs keep the technician and the manager working on the same record.
Abnormal reading
A reading over its limit is flagged on the spot, becomes a linked incident, and turns into corrective work once the manager approves it.
Rejected checklist
The manager rejects a checklist with a written reason, and it goes back to the technician to fix and resubmit.
Outcome
I delivered iPad and web mockups for both roles and revised them after the host company's review. This is a concept: it was not built or tested with users. The team's roadmap proposed a pilot at one plant as a later phase.
Reflection
what I learned
I designed from the client's answers, not from watching technicians at work, and nothing was tested with real users. So the assumptions behind QR scanning, blocked submission and field conditions like poor light or patchy network still need validation, ideally in a pilot at one plant.
Student research project on sustainable reverse logistics
Sustainable Returns App
A mobile concept for handling returns from several marketplaces in one place, with greener packaging and transport built into the return.
Overview
- Role
- UX/UI Designer in a five-person research team
- Timeline
- October 2024 to March 2025
- Platform
- Mobile app, in Vietnamese
- Status
- Concept, not built or tested
- Sample data
- Illustrative, no marketplace affiliation
My role
- I designed the app interface after the analysis.
- The survey, statistics and recommendations were team work.
Research
The study asked how green logistics and reverse logistics affect online shoppers' satisfaction with returns in Ho Chi Minh City.
Survey
337 valid responses to an online survey, December 2024 to February 2025.
Sample
A convenience sample, mostly students aged 18 to 22.
Analysis
Reliability checks, factor analysis and regression.
Expert input
A logistics lecturer reviewed the research model over two rounds.
Findings
Significant predictors
Green transport, green packaging, return cost and customer service.
Not significant in the regression model
Return processing time and return policy.
How much it explains
Together the factors explained about a fifth of the variation in satisfaction.
The design problem
Take a shopper from reporting a problem to receiving a refund in one flow, with the greener choices the research pointed to built in.
What the survey did not tell us
It measured satisfaction with returns, not how people would use an app.
From findings to concept
01
Build the green choices into the return form
- Problem
- Green packaging and green transport were among the factors linked to satisfaction, so the return needed a place for those choices.
- Decision
- The return form asks for a packaging type, such as biodegradable, and a return method. Doorstep pickup adds a green vehicle choice shown as three large tiles.
- Trade-off
- The return form gets two extra choices to make.
- Why it matters
- Shoppers see the greener options while arranging the return, without going to a separate page.
02
One place for returns across marketplaces
- Problem
- Shoppers buy across several marketplaces, each with its own return flow (our design assumption, not a research finding).
- Decision
- One list with Orders and Returns tabs shows every return with a status label. A filter narrows it by marketplace and by stage, from pending to refunded.
- Why it matters
- A shopper can find any return without switching apps.
Final solution
One return, from reporting the problem to getting the money back.
Request a refund
The shopper says what went wrong, adds proof and sees the refund amount before sending the request.
Send it back and track it
The shopper chooses how to hand the item over, then follows the return until the refund arrives.
Reflection
The concept was proposed in the team's research report in March 2025. It was not built or tested.
what I learned
The survey showed which factors were linked to satisfaction, not how shoppers would use a returns app, and its sample was mostly students. Before building, I would test whether shoppers pick one app over each marketplace's own flow and whether they notice the green options.
Business Website Development course project
ClayCo
An e-commerce site for a fictional ceramics brand, where a personalised order stays intact from the product page to the store's admin.
Overview
- Role
- UX/UI Designer and Frontend Contributor
- Team
- 5 people
- Timeline
- Submitted November 2024
- Stack
- Figma, Angular, NodeJS, MongoDB
- Brand
- Fictional course concept
My role
- I contributed to the Figma design and to part of the frontend.
- Teammates also designed and coded, and the business analysis was team work.
Business concept
A fictional brand our team created for the course, selling personalised handmade ceramics alongside plants and flowers.
Customers
Craft lovers and eco-conscious shoppers, defined in the team's business model.
Baseline
A review of three competitor websites.
The design problem
Sell a custom request as easily as a standard product, and keep it attached to the order all the way to the store's admin.
How we built it
- 1
Frame the business
Team effort
Business model, SWOT and the competitor review.
- 2
Map the flows
Team effort
Use cases and process maps for customers and the admin.
- 3
Design
I contributed
Shared visual rules, wireframes, then high-fidelity mockups.
- 4
Build
I contributed
The customer site and admin area, coded with Angular, NodeJS and MongoDB.
Design choices
01
Make personalisation a step on the product page
- Problem
- Personalised pieces were the brand's core offer, but a standard product page only offers Add to cart.
- Design choice
- Customisable products show Customize next to Add to cart. It opens a panel for placement, text, an optional image and extra details, then adds the item to the cart.
- Why it matters
- The request travels with the item, so the cart shows the engraving right under the product name.
02
Support the three payment paths in the checkout flow
- Problem
- The team's order flow defined three ways to pay: cash on delivery, internet banking and MoMo.
- Design choice
- Checkout offers all three, and MoMo opens a Scan to Pay screen with the order number and amount. A note field under the delivery details holds gift messages and delivery instructions.
- Why it matters
- Every payment path has a place in checkout, and instructions travel with the order.
03
Keep the customer's request visible on the admin side
- Problem
- After checkout, the store works with the order from the admin area, not the customer's screens.
- Design choice
- The admin order detail shows the payment method, items, buyer details and order note in one view. A status dropdown under the order handles updates.
- Why it matters
- The engraving reaches the store intact, and the admin updates the order from the same screen.
Final solution
One personalised order, followed from the product page to the store.
Place a personalised order
The shopper engraves a mug, checks the cart and pays with MoMo.
After checkout
The order reaches the admin with the engraving in its note, and the shopper can follow delivery.
Reflection
The team submitted mockups and a coded site in November 2024, with favourites, reviews and promotions left for later.
what I learned
The design came from the team's business goals and a competitor review, and no shoppers tested it. I would start by checking whether shoppers find Customize and can describe what they want with text and an image alone.