Product Designer at Charles Schwab
Yashas H Mohana
I design financial and enterprise products where the hard part is helping someone decide, not helping them click.
Selected work
Three case studies. Click one to open it.
The pause before you move money
Redesigning MoneyLink account funding so customers know which account they're using, whether it's ready, and when their money lands.
Read case studyOne way to do the same thing
A shared set of interface patterns, designed and built, for eleven enterprise modules that each did things their own way.
Read case studyAn AI people didn't believe
Research, a name and a visual identity that took trust in an AI assistant from 2.4 to 4.6 out of 5.
Read case studyAbout me
I studied architecture before interaction design, and the habit stuck: before changing anything, I want to know how people move through the structure today, and where they stop.
I work from evidence. Behaviour data shows where something's wrong, conversations explain why, testing proves whether the fix worked.
I also write front-end code, so I bring engineers in early and design the pending and failed states, not just the happy path.
AI financial decision workspace
40+ interaction states for planning, confidence, citations, human approval and undo.
Enterprise operations copilot
What an agent may do alone, and what needs a person to approve it.
Product and strategy
- Discovery
- Problem framing
- Journey mapping
- Trade-offs
- Success metrics
Research
- Interviews
- Usability testing
- Heuristic review
- Funnel analysis
- Synthesis
AI and agentic UX
- Human-in-the-loop
- Trust and transparency
- Uncertainty
- Recovery
- Audit trails
Craft and systems
- Interaction design
- IA
- Design systems
- WCAG 2.1 AA
- React, HTML, CSS
Got a hard decision to design?
Financial, enterprise or AI products where the choice is the difficult part. I'd like to hear about it.
The pause before you move money
Customers could reach the right screen and still not know whether to press the button.
- Company
- Charles Schwab
- My role
- Product designer
- Team
- PM, research, data, eng, content, compliance
- Timeline
- Jul 2025 — now
Screens are my own re-drawn wireframes. Names, numbers and some flow details are changed to respect confidentiality.
It wasn't a transfer problem. It was an understanding problem.
A lot of what got logged as "transfer issues" happened to people who did everything right — and still stopped, unsure. I mapped the flow as the questions customers were asking, and marked where each one went unanswered.
Fig 1.1 Red marks the question customers were left holding. Green is what the redesign put there.
Data found the pauses. People explained them.
With our analyst: where people paused, went back, or left.
Plus support themes, to learn why the pauses happened.
Journey mapped in FigJam around what customers ask.
Realistic tasks, watched for hesitation, changed the screens.
Every screen answers one question.
Status before selectionAn account that isn't ready says so, and says when.
Timing beside the numberThe arrival date moves up from the last screen.
Terms kept, not buriedRequired disclosure folds to one line you can open.
A promise, not a spinnerThree stages, each with a date the bank can keep.
Fig 1.2 Mid-fidelity flow. Swipe on mobile.
The system doesn't always know yet. The screen shouldn't pretend it does.
I brought engineers in early to learn what arrives late, then designed those moments as states — not errors.
Still verifying
Ready to use
Didn't go through
The constraint I couldn't design away
I worked line by line with compliance. I could change how and when a condition appeared — not remove it. Knowing that early kept every argument about the part I could change.
What changed
Reported for the money-movement work this project sits inside.
One way to do the same thing
Eleven modules built by different teams. A failed form could fail three different ways, depending on where you were.
- Company
- Mphasis, for an enterprise client
- My role
- UX engineer and designer
- Team
- 5 Agile squads, product owners, QA
- Timeline
- 2022 — Aug 2023
Client unnamed. Wireframes are re-drawn with placeholder data to respect the client agreement.
Nobody designed the inconsistency. It just added up.
Each team solved search, tables, forms and errors when they needed to. Every answer was reasonable. Together, employees relearned the same task in every module.
Fig 2.1 The same failed submission in three modules, and the pattern that replaced all three.
Audit, define, build, roll out.
Differences grouped by the problem they caused people.
Plus the exact button and status wording.
Built with developers, reviewed in GitHub.
Usage notes in Confluence, tracked in Jira.
Designed once, with the edge cases in.
Fig 2.2 Table pattern: search, filters that show they're on, status in words and colour.
Fig 2.3 Form pattern: the error sits on the field, and the summary says nothing else was lost.
Every pattern shipped with its awkward states
Loading
No results
No claims match "Type: Claim"Clear filterLong text
Request to update the mailing address for the primary policyholder following relocation to a new state with different coverage rulesShow moreNo permission
View onlyYou can see this request. A supervisor can approve it.Failed
Fig 2.4 The five states each pattern had to cover before a squad could use it.
One screen didn't fit. So it didn't have to.
Fig 2.5 Familiar controls, plus the context this decision needed. Figures are placeholders.
Why keep it
A standard confirmation couldn't hold what the approver needed to see. Forcing the pattern would have made a hard decision harder. A system that can't bend gets forked quietly — then you have eleven exceptions and no system.
Fig 2.6 Simplified adoption view across the eleven modules.
What changed
From the client's delivery reporting.
An AI people didn't believe
Ultopia scores six areas of your life and uses AI to tell you what to do next. People read the advice. They didn't trust it.
- Product
- Ultopia, iOS lifestyle app
- My role
- UX researcher and designer
- Team
- CEO, product, engineering
- Shipped
- 2025, App Store
The brief was a redesign. The problem was trust.
Early feedback said people used the app and ignored the AI it was built around. Before touching screens I set out to find why: a competitive scan, a heuristic review, then usability tests.
48 issues across 28 screens.
Against Nielsen's ten heuristics.
Plus 18 moderate and 17 minor.
Every severe and moderate issue covered.
To learn which issues truly stopped people.
Wait, I can't see what I'm typing in the password box.
Participant 1, create accountIt's telling me to pay off my 60 dollar debt. I think it meant 60 thousand.
Participant 4, AI recommendationHonestly it looks like random text, not AI. Doesn't feel like ChatGPT.
Participant 6, discipline tracking
48 issues, three real problems.
Technical blockers
Keyboard hid the password. Zero wasn't accepted. Sharing dead-ended.
AI trust erosion
Wrong maths in public, and output that looked like plain text.
Navigation confusion
Logo used as a menu. Text-only lists. Look-alike labels.
Giving the AI a name.
Naming is the first step to treating something as an entity rather than a function. I derived Ultima from Ultopia so it felt native — the same move as Google and Gemini.
ULTOPIA
shared "U" root
Ultima AI
What intelligence looks like, elsewhere
Fig 3.6 Blue says reliable; purple and pink say generative. Everyone had moved from robot icons to fluid forms.
From robot to orb.
Mid-fidelity prototypes
Fig 3.10 Three directions tested for transparency, depth and rhythm.
Final visual signature
Making AI insight look like insight.
Each content type in its own container.
No more scrolling inside a tiny box.
Headings and titles stop looking alike.
The orb tags everything the AI wrote.
Navigation
Same labels, same order — an icon on each row turns reading into scanning.
What changed
Post-update usability study against the original build, same tasks.
Reflection
Leadership later renamed Ultima to YUVA — Your Ultopia Virtual Assistant. The argument I made, that the AI needed an identity at all, is the one that won; the name belonged to the company. The interface, visual language and most of the 48 fixes still ship.