Ayush Rajput — product designer, UX researcher, creative technologist. I design the human layer of complex systems.

I design the human layer of complex systems.

Works

Every product here starts from one question: who does this help, and what's in their way?

2026 · Selected work
01Fintech · Product design
For
International students and newcomers learning the US financial system.
Problem
“I can find a list. I just don't know which list applies to me yet.”
Impact
Tested with six newcomers, then rebuilt in three phases around what they asked for.
View case study →
02Civic tech · UX research
For
Analysts in Prince George's County, Maryland.
Problem
Maintenance gaps hide across 311 requests, sidewalk and road-condition data.
Impact
A concept map that turns three disconnected county datasets into one ranked repair list.
View case study →
03B2B SaaS · Product design
For
Designers, researchers and engineers handing off work in progress.
Problem
When work changes hands, the decisions and open questions get lost.
Impact
Shaped by my own interviews and survey; a live prototype anyone can try.
View case study →

Play

Side quests: experiments and games I make to test an idea. Hover a flower to let it grow.

Off the clock · 02
↑ hover to grow

The Voice of ✨

Experiment · Accessibility

The ✨ sparkle tells sighted people “this is AI”. What does it sound like to someone using a screen reader? Non-visual signifiers for AI.

↑ hover to grow

The Security Deposit

Game · Horror-comedy

A browser game about moving out of a haunted apartment and trying, against all odds, to get your deposit back.

Say hi

Hiring, collaborating, or just curious? I'd love to hear from you. Let's check our calendars →

The human · 01
↑ hover me

I’m Ayush. Visual designer turned HCI researcher, which means I used to argue about hex codes and now I argue about mental models. I design for the first ten seconds someone spends with a system, the split second where it either clicks or someone sighs at their screen.

Now
M.S. HCI, University of Maryland
Before
Product & Digital Designer, VGC Mumbai
Looking for
Summer 2027 internships

Tools I reach for. Some for pixels, some for code, all for thinking out loud.

The toolbox · 25

Hover a tile to see its name. Hover the row to pause it.

Let's talk: book a call

Pick a time that works for you. Thirty minutes, no slides, just a conversation.

Calendar · 01

Coffee chat with Ayush

Hiring for a design or research role, working on a tricky product problem, or just curious how I think? Grab 30 minutes and let’s talk.

  • 30 minutes
  • Google Meet
  • Your time zone
01ArrivalLANDING · Case studyDay 01 / 90
LANINGLANING
Role
Product designer · research, UX, UI, prototype
Evidence
8 interviews · 20 surveys · 22 sources · 6 usability sessions
Output
Responsive web prototype

A calm guide to the first 90 days of US money.

02The problemLANDING · Case studyDay 02 / 90

Research question

How can a new international student tell which US money tasks apply to them, in what order, and who decides each one?

Hypothesis

If the first 90 days are laid out as one journey, ordered by real dependencies, with a plain explanation and the deciding source beside every step, students will act with more confidence without sharing any personal or financial data.

PrimaryInternational students in their first 90 days

No SSN, no US credit, no US bank history. Six institutions to deal with, in no shared order.

Why it matters2.85 / 5

average confidence about which tasks applied to them, across 20 surveyed students.

Demand16 of 20

would use or consider a setup companion.

03ContextLANDING · Case studyDay 03 / 90

One person.
Six deciders.
No shared order.

Youthe only one who sees all of it
Bank or credit unionWhich ID it accepts, fees, overdraft
LandlordHow rent is paid, when, late fees
Your universityThe bill, its date, payment methods
International officeSSN vs ITIN, Form 8843
Transfer providersFee and exchange rate
State or countyWhen a deposit comes back

Dashed lines: one step waits on another. Hover a decider.

04The pileDay 07 / 90

Everyone hands you a list.

Nobody says which one applies.

  1. 01Proof of address
  2. 02SSN
  3. 03Lease
  4. 04Bank account
  5. 05Phone plan
  6. 06I-20
  7. 07Debit card
  8. 08Security deposit
  9. 09Utilities
  10. 10Money transfer
  11. 11Health insurance
  12. 12Rent
  13. 13Credit card
  14. 14University bill
  15. 15Credit history
  16. 16Scam calls
05Desk researchLANDING · Case studyDay 10 / 90

22 sources. Everyone sells a product or explains a concept.

22 of 22 sources read

13 government5 university4 provider pages
Who people useGood atWhere it stops
Wise, RemitlyFast, transparent transfersTransfers only
SelfCredit-builder loansAssumes you know US credit
Bank student guidesAuthoritative for one bankNot other banks, rent, school
CFPB resourcesUnbiased educationDense, not personal
University officesLocal, specificThin on money topics
LANDINGWhat applies, in what order, who decidesNever sells or ranks
06ListeningLANDING · Case studyDay 14 / 90
0

Surveys · plus 8 interviews
8 home countries

0/5 confidence about what applies
0/5 clarity about what to do next
16/20would use or consider a setup companion

Biggest difficulty

Requirements4
Task order4
Terminology4
Waiting on others3
Trusting sources3
Managed fine2

What they wanted · no single winner

Checklist4
Single next step4
Plain explainer4
Timeline3
Search3
Comparison2
07Evidence limitsLANDING · Case studyDay 16 / 90

What this research can't tell us.

  1. 01Small, purposive sample

    28 people through university networks. Strong signals, not estimates.

  2. 02Self-reported

    Confidence as remembered, not observed behaviour.

  3. 03One synthesiser

    No second coder checked the themes.

  4. 04Other schools' guidance

    Borrowed from four schools; your office is the authority.

  5. 05Sources age

    9 of 22 re-read in Oct 2026; the rest are labelled.

  6. 06No outcome data

    Nothing yet measures faster setup.

08SynthesisLANDING · Case studyDay 30 / 90
  1. 01Know what applies
  2. 02One next step
  3. 03Short version first
  4. 04Remove pressure
  5. 05Privacy first

Finding

Long checklists, no way to tell which items were theirs.I01, I04, I08 · S01, S07, S19, S20

Counterevidence

Confident users navigated university portals on their own.I06, S06

So the design

Gates nothing. Questions can be skipped; any task can be marked not applicable.
09People and weeksLANDING · Case studyDay 45 / 90

Three people.
Twelve weeks.

AUncertain Starter

Needs one next action and the official link.

BCapable Newcomer

Needs a non-linear journey; anything skippable.

CCautious Organizer

Needs local-only storage, no credentials ever.

Days 0–14Establish

Unclear document rules

Days 15–30Connect

Delays outside their control

Days 31–60Understand

ACH, APR, fees, terms

Days 61–90Routines

Timeline pressure

0153045607590

Friction peaks · days 15–60

Days in the US

Day 1
10Product boundaryLANDING · Case studyDay 50 / 90

Guidance,
not a bank.

The line was drawn before the first screen and held through every round.

LANDING does

  • Shows which tasks may apply
  • Orders tasks by real dependencies
  • Explains terms, with the source
  • Gives tools to prepare
  • Names who decides each step

LANDING never

  • Opens accounts or moves money
  • Guarantees eligibility
  • Ranks or recommends providers
  • Asks for SSNs, logins, income
  • Asks about immigration status

Seven design principles, from the research

  1. 01Ask before tellingShort, skippable questions; never status, IDs or money.
  2. 02One next step, honest about waitingLead with one action; say what blocks it.
  3. 03Short version first, source alwaysPlain language on top, the deciding source beneath.
  4. 04Say who decidesEvery task names the institution that owns it.
  5. 05Teach the factors, never pickExplain what to compare; no rankings.
  6. 06Nothing sensitive, everNo account, no linking, nothing identifying.
  7. 07Every step ends somewhereEach action has a place in the product. Added after the audit.
11System designLANDING · Case studyDay 55 / 90

15 tasks.
One model.

Compare banksOpen accountAccount terms AutopayRent only if your landlordneeds a US account

Worked out by LANDING

ReadyNot startedNeeds verification

Set by you

In progressWaitingCompletedDeferredNot applicable
12PrototypeLANDING · Case studyDay 60 / 90

Three flows,
on real screens.

Try the prototype
Flow 1 · Does this apply → a bank, recording of the prototype
01From "does this apply?"
to a bank you can apply to
Flow 2 · One payments list, recording of the prototype
02One payments list
that feeds five tasks
Flow 3 · When it isn't up to you, recording of the prototype
03When the next step
isn't up to you
13Usability round oneLANDING · Case studyDay 70 / 90

6 moderated sessions · October 2026

It worked as a map.
Not yet as a guide.

Worked

6/6praised no sign-up, password or bank login
✓Sequencing, prerequisites and status were easy to follow

Didn't

4/6had scam calls; the product didn't cover them
×"Next action" said check or find out, then sent people to Google
×No way to compare options side by side

"When I'm sitting in 'Waiting' for my debit card, the app is just blank."

P2

"I want to know what factors I should be comparing, but please don't just tell me which bank."

P5

"It just tells me to go look up fees. I was hoping it would explain what a checking account is."

P3
14Fix one · organiser to guideLANDING · Case studyDay 75 / 90

Fix one: from organiser to guide.

Round-one findings → features · every new fact checked against an official or university source

It pointed instead of teachingWhat to know cards
Let me compare, don't tell meA comparison sheet you fill in
4 of 6 got scam callsScam red flags, first month
P2: the blank waitWhile you wait
Transfers were guessworkA transfer calculator
15Fix two · guide to toolLANDING · Case studyDay 80 / 90

Fix two: from guide to tool.

9/14tasks ended in an action with nowhere to do it
5questions whose answers changed nothing
6more dead ends outside the tasks
Before · a "Yes" that changes nothing
After · payments you fill in, recording of the prototypeAfter · payments you fill in
After · every step ends somewhere

From a design audit, not new participant data. The next round is where it gets checked.

16Next roundLANDING · Case studyDay 85 / 90

What round two
has to prove.

Six participants · targets set in advance, not results

needs to pass may miss

  1. Fix oneWhat to know

    Explain checking vs savings in their own words

    4 of 6
  2. Fix oneComparison sheet

    Add two banks and fill the factors that matter

    4 of 6
  3. Fix oneScam safety

    Read a sample call script, decide what to do

    5 of 6
  4. Fix oneWhile you wait

    Find something to prepare while a card is in the mail

    4 of 6 in under a minute
  5. Fix twoPayments list

    List two payments, then find them on the journey

    4 of 6
  6. Fix twoEvery step ends

    Finish three tasks without leaving LANDING

    4 of 6
  7. AuditAccessibility

    Core flows by keyboard and screen reader at 320px

    WCAG 2.1 AA
    0 critical issues

What the design adds

A guidance layer, not another product.Status that explains itself.One list as connective tissue.
17ReflectionLANDING · Case studyDay 90 / 90

What I learned

A clearerstart.

Testing changed what LANDING was for. Round one showed it worked as a map but not yet as a guide, so I rebuilt it twice: from organiser to guide, then from guide to tool. Every step now ends somewhere, and nothing ever asks for a credential.

Validate

Round two, six participantsPass marks set before testing

Strengthen

A second coder on the five themesRe-check all 22 sources

Take it further

Keyboard and screen reader at 320pxMeasure whether setup gets faster
Next case studyCivic Signal
01ArrivalCIVIC SIGNAL · Case studyLead 01 / 18
0 — 1 mi38.83° N · 76.85° W

CIVIC
SIGNAL

A map-first investigation tool for Prince George's County, Maryland. It puts what residents report next to what the county measures.

Role
Solo · research, data, UX, UI
Timeline
4 months
Data
69,461 requests · 20,766 roads
Tools
Python · HTML · Canvas
02The questionCIVIC SIGNAL · Case studyLead 02 / 18
0 — 1 mi38.83° N · 76.85° W

Research question

Where do Prince George's County complaints and road data disagree, and how can anyone check responsibly?

Hypothesis

If both are shown together, with their limits beside every number, non-GIS staff can find credible leads quickly and turn them into specific questions for the county.

Primary · proposedCouncil constituent-services staff

Hear about roads daily, answer to residents, can request records. Spreadsheets yes, GIS no.

SecondaryCivic analysts and local journalists

Need traceable numbers and clear caveats before publishing.

03Two recordsCIVIC SIGNAL · Case studyLead 03 / 18
0 — 1 mi38.83° N · 76.85° W
County roads by pavement score
Roads with 311 requests

The county measures.

Pavement scores, 0–100. 20,766 segments. Survey date unknown.

Residents report.

311 requests. 69,461 of them, in two systems.

Drag the line to compare.

04Secondary and practitioner researchCIVIC SIGNAL · Case studyLead 04 / 18
0 — 1 mi38.83° N · 76.85° W

A quiet street is not
a good street.

Kontokosta & Hong, 2021

311 behaviour varies with neighbourhood demographics.

Compare rates, never rank by totals.
O'Brien, Sampson & Winship, 2015

A request measures conditions and the willingness to report.

"Poor and quiet" gets its own lead type.
ASTM D6433

PCI rates visible road-surface distress.

Label PCI as road surface only.

Who I still need to hear from

DraftedData-field questions for DPW&TWhen PCI was surveyed; what codes mean
Not yet3–5 council constituent-services staffHow complaints reach them today
Not yet1–2 reporters or civic analystsWhat makes a public-data claim publishable
05Source auditCIVIC SIGNAL · Case studyLead 05 / 18
0 — 1 mi38.83° N · 76.85° W

Before designing anything, I audited every source.

20162018202020222024202620282030
System switch · early 2022
County Click 311 · 33,999Half lack coordinates
PGC311 · 35,462No close dates
PAMS pavement · 20,766Survey date not published
PAMS projects · 1,910Thins out after FY2024
Sidewalks · 60,029PCI copied from road
Road centerlinesNo condition data
06Affinity mapCIVIC SIGNAL · Case studyLead 06 / 18
0 — 1 mi38.83° N · 76.85° W

25 observations. Five questions the data couldn't answer.

Where exactly?

8.4% name only a street

Pre-2022: half lack coordinates

4,052 on state roads

Block-level addresses only

How old is it?

No PCI survey dates

Projects thin after FY2024

364 projects stuck "approved"

No close dates in 311

Who is responsible?

Codes 4 and 5 undefined

Classes P and S undefined

State highways not in PAMS

Districts from a road attribute

Who reports?

Reporting varies with demographics

2,103 poor segments, no requests

Bigger districts, more requests

A request signals care

What changed?

Systems switched early 2022

Emergencies tripled: 541 → 1,755

Categories merged after 2023

Request IDs repeat

07Journey and workflowCIVIC SIGNAL · Case studyLead 07 / 18
0 — 1 mi38.83° N · 76.85° W

Today · proposed current-state journey

Friction ■ per stage

Hears a complaintMotivated

No map of requests over time

Looks for evidenceSlowed down

Searches one request at a time

Tries to compareStuck

Can't join data without GIS

Asks the countyUnsure what to ask

Sends anecdotes, not evidence

Reports backExposed

No traceable evidence to share

Target workflow · each step answers the stage above it

1Scan

Where signals agree or disagree

2Narrow

A district, by rate, vs the county

3Verify

A lead and what it can't show

4Record

Save evidence, label neutrally

5Ask

Specific questions to the county

08Data preparationCIVIC SIGNAL · Case studyLead 08 / 18
0 — 1 mi38.83° N · 76.85° W
69,461

311 requests, from both systems

  1. Merge69,461requests from both 311 systems
  2. Window36,536since 2022, locations reliable
  3. Match30,750within 60 m, same street name
  4. Set aside5,786off-network, low confidence
  5. Robust456Pattern C, false positives removed
09Spatial analysisCIVIC SIGNAL · Case studyLead 09 / 18
0 — 1 mi38.83° N · 76.85° W

Three first looks, and a warning about time.

Fig. 1 21% of road length scores Poor. Median PCI 66.
Fig. 2 Requests concentrate inside the Beltway. Density is not a ranking.
Fig. 3 Where the two signals diverge, before robustness checks.
Fig. 4 Before the 2022 switch, under half of requests carry coordinates, so the tool defaults to "Since 2022".
10Candidate investigationsCIVIC SIGNAL · Case studyLead 10 / 18
0 — 1 mi38.83° N · 76.85° W

Where the records disagree.

Pattern A, all segmentsPattern A, strong leads only
A2,103

Poor, few or no requests. Who reports?

Pattern B, all segmentsPattern B, strong leads only
B511

Good score, heavily reported. Stale score?

Pattern C, all segmentsPattern C, strong leads only
C681

Poor, reported, no work on record.

Removed: segments under 100 m · PCI of exactly 0 · roads where most requests name only the street

11Product strategyCIVIC SIGNAL · Case studyLead 11 / 18
0 — 1 mi38.83° N · 76.85° W
Staff need a way to test residents' road complaints against the county's own records, because today they forward anecdotes without evidence.
  1. HMWShow where complaints and scores disagree, without implying either is the truth?
  2. HMWMake a caveat as visible as the number it qualifies?
  3. HMWTurn a pattern on a map into questions the county can answer?
PriorityRequirementStatus
MustMap of roads by PCI with 311 requests, filterableBuilt
MustDistrict vs county, by rateBuilt
MustRanked leads with location warningsBuilt
Must"What this can't tell you" on every roadBuilt
ShouldNotebook that drafts a neutral summaryBuilt
ShouldTimeline with the 2022 switchBuilt
ShouldHigh-contrast themeBuilt
CouldOfficial boundaries, equity layerNext
12IA, flow, wireframesCIVIC SIGNAL · Case studyLead 12 / 18
0 — 1 mi38.83° N · 76.85° W

Six rules, one structure.

  1. 1Leads, not verdictsNever "neglected".
  2. 2Caveats beside numbersNot on a methods page.
  3. 3Rates over totalsPer km per year.
  4. 4Every number traceableObserved or Derived.
  5. 5Map first, table alwaysA list behind every map.
  6. 6Accessible by defaultWCAG 2.1 AA floor.
Civic Signal
Map

Area profile

Ranked leads

Timeline

Requests

Searchable table

Location quality

Roads

Segments

Pattern filter

Projects

Status list

Timeline

Areas

District compare

Evidence summary

Methods

Sources

Definitions

Main flow · a complaint becomes a question for DPW&T

1Scan countyMap
2Pick District 6Areas
3Open top leadMap
4Read limitsMethods
5Check shortlistRoads
6Save to notebookAreas
7Copy summary to DPW&TAreas
13Design explorationCIVIC SIGNAL · Case studyLead 13 / 18
0 — 1 mi38.83° N · 76.85° W

Seven versions to get closer to how GIS tools are built.

The people I designed for already read places on maps, and the tools behind county data share a grammar: a docked layer panel, a full-bleed map, a legend for every colour. Borrowing that grammar meant less to learn and more trust in the numbers, so attention stays on the evidence, not the interface.

V1Dark analystDropped

Read as a specialist tool; dark ground blurred the PCI blues.

V2Light, filter barDropped

Eight dropdowns made filtering the main event.

V5Floating cardsDropped

Every element a rounded card; looked like my other work.

V6Map editionChosen

One docked panel, full-bleed map, square corners.

14Final experienceCIVIC SIGNAL · Case studyLead 14 / 18
0 — 1 mi38.83° N · 76.85° W
District 6 roadsRitchie Rd highlighted in orange
District 6
Guided walkthrough, recording of the prototypeGuided walkthrough · 12 steps
Try the prototype
15Final experienceCIVIC SIGNAL · Case studyLead 15 / 18
0 — 1 mi38.83° N · 76.85° W

Complaint in. Question out.

Map, recording of the prototypeMap
1MapDistrict flies into focus; the panel compares it with the county.
Lead detail, recording of the prototypeLead detail
2Lead detailA lead check, requests by year, and what it can't show.
Road shortlist, recording of the prototypeRoad shortlist
3Road shortlistEvery lead in a sortable table, with a map card.
Evidence notebook, recording of the prototypeEvidence notebook
4Evidence notebookA neutral summary that copies into an email.
16EvaluationCIVIC SIGNAL · Case studyLead 16 / 18
0 — 1 mi38.83° N · 76.85° W

Reviewed against Nielsen's ten.
Eight fixes.

Consistency

Five control heights and radii

Two heights, one radius

Affordance

A status pill looked like a button

A small white tag

Visibility

Filters cut off below the fold

A four-column panel

Conflicts

Filters and Layers overlapped

One panel at a time

Purpose

Three map styles, for looks

View, Theme, Detail

Error prevention

Weak leads looked strong

"Strong leads only", on

Aesthetic

Mono font mixed with UI font

One typeface: Helvetica

Alignment

Toolbars off the grid

Edges follow the columns

Accessibility audit · contrast

Body text8.76:1 pass
Map labels, dark3.51 → 6.91:1
Good-road blue1.48 → 1.72, mitigated
Request dots2.70:1, mitigated
17Usability plan and iterationsCIVIC SIGNAL · Case studyLead 17 / 18
0 — 1 mi38.83° N · 76.85° W

Next: five people,
five tasks.

Planned · not yet run

T1How does District 6 compare?States the rate, not the total
T2Find the strongest lead, say whyNames Ritchie Rd with two pieces of evidence
T3Is Walker Mill Rd a strong lead?Notices the street-name-only warning
T4Save the lead, prepare a noteUses the notebook, copies a summary
T5Make the map easier to readFinds High contrast or Less detail
V1Dark dashboardV2Light, filter barV3Card workspaceV4Real basemapV5Floating toolsV6Map editionV7Strong-lead filter
18ReflectionCIVIC SIGNAL · Case studyLead 18 / 18
0 — 1 mi38.83° N · 76.85° W

Leads,
not verdicts.

Designing for uncertainty became the core of the work. The data audit showed raw totals mostly measured who uses 311, so every number carries its caveat, findings are leads to check, and every investigation ends in questions for the county.

Validate

Run the five-task testInterview 3–5 council staffHand-check 100 matches

Strengthen

Official district boundariesPopulation rates, hotspot test

Take it further

Publish to ArcGIS OnlineRebuild with the ArcGIS Maps SDK
Next case studyRelay
relayCover
0119

Case study No. 03 · Product design · B2B SaaS · 2026

relay

The work moves.
The why stays.

A handoff tool for product teams. When work changes hands, the decisions, the reasons behind them and the open questions travel with it.

The Relay Return Path mark in green glass
48people surveyed
6interviewed, both sides
7issues found, all fixed
3looks before the final one
relayOverview
0219

Overview

Product teams hand work over all the time. The Figma file and the ticket move with it. Why a decision was made, what's still a guess and what happens next usually stay in a meeting. Relay keeps them with the work.

Role
Product designer, solo
Timeline
10 months
Type
B2B SaaS · web, desktop, mobile
Methods
Competitive analysis, survey N=48, interviews n=6, synthesis, IA, wireframes, prototype, expert review, critique
  1. 04DiscoverTools, plan, interviews, survey
  2. 08DefineAffinity, journey, archetypes, problem
  3. 12IdeateOpportunities, priorities, concepts
  4. 14DesignFlow, wireframes, prototype
  5. 16TestExpert review, critique rounds
  6. 18DeliverThe product, and what's next
relayThe problem
0319

The problem

0.0hours

a week, on average, spent rebuilding context on unfinished work.

70.8%name unclear decision rationale as a top handoff problem
18.8%say the original rationale is easy to find
72.9%often ask a teammate to re-explain a meeting decision

The files survive the handoff. The reasoning doesn't.

relayDiscover
0419

01 · Secondary research

Every tool keeps the artifact.
None keeps the handoff.

The five tools that came up most in a product team's day, held against what the person inheriting the work needs.

Designed for itPartly, inside its own wallsNo

What this meant. Don't replace these tools. Link to them, and add the layer none of them has.

Figma
Dev Mode
LinearConfluenceSlack
canvas
LoomRelay
The why behind each decision
How settled each decision is
Open questions stay with the work
A next owner who accepts
A summary written for the receiver
Context from more than one tool
Needs fully metPartly met040303040460
relayDiscover
0519

02 · Research plan

Measure the problem,
then understand it.

  1. Q1What must a receiver figure out before continuing?
  2. Q2How do teams preserve why a decision was made?
  3. Q3What happens to open questions and blockers?
  4. Q4How do people learn the next action and its owner?
  5. Q5How do people move between Slack, Figma, Linear and Notion?
  6. Q6What belongs in a handoff, and what stays in existing tools?
relayDiscover
0619

03 · Interviews · six people, both sides of the handoff

P01
"I end up writing long Slack posts explaining what changed."
Product designer · sender

Fed intoTheme 1 · The why gets lost

P02
"If an edge case isn't drawn, I usually have to pause my ticket."
Front-end engineer · receiver

Fed intoTheme 2 · Gaps are ambiguous

P03
"The decision makes sense if you were in the meeting."
Product manager

Fed intoTheme 1 · The why gets lost

P04
"A finding can turn into a requirement surprisingly quickly."
UX researcher

Fed intoTheme 2 · Gaps are ambiguous

P05
"I don't need their entire project history."
Cross-project engineer · receiver

Fed intoThemes 3 and 4 · A starting point, scattered context

P06
"A handoff isn't complete just because somebody posted a link."
Design lead

Fed intoTheme 3 · Receivers need a starting point

relayDiscover
0719

04 · Survey · N = 48

Hours a week rebuilding context

Product management5.2 h
Engineering4.8 h
Product design3.9 h
All respondents4.6 h

Top friction when receiving work

Unclear decision rationale70.8%
Ambiguous edge cases60.4%
Information across 3+ tools43.8%
Next owner or step not stated25.0%

Confidence at handoffs · 1 disagree to 5 agree

The original rationale is easy to find

2.35 · 18.8% agree

I know whether a missing UI state was intentional

2.12 · 14.6% agree

Our tools preserve decision context

2.48 · 22.9% agree

I often ask a teammate to re-explain a decision

3.92 · 72.9% agree
relayDefine
0819

05–06 · Affinity map → themes → insights

InterviewSurvey

Theme 1

The why gets lost

Which version was agreed after review isn't recordedP01
Decisions make sense only to people in the meetingP03
Trade-offs from huddles evaporateP03
70.8% name unclear rationaleSurvey

Receivers don't need more documents. They need the reason attached to each decision.

Theme 2

Gaps are ambiguous

Can't tell "rejected" from "nobody got to it"P02
A missing edge case means pausing or guessingP02
Research nuance disappears in requirementsP04
60.4% name ambiguous edge casesSurvey

Silence reads as a decision. Say what's settled, a guess, or left out.

Theme 3

Receivers need a starting point

Want what's done and blocked, not the historyP05
Posting a link isn't a finished handoffP06
Links sit unopened until sprint planningP06
25.0% say the next owner isn't statedSurvey

A handoff is complete when the receiver can act, not when the sender hits send.

Theme 4

Context is scattered

Four Slack channels, two tickets and a PRP05
The same question in Slack and on the canvasP01
Three Notion docs to update per decisionP03
4.6 h a week rebuilding contextSurvey

Point into the tools people already use, and summarise around them.

relayDefine
0919

07 · Inheriting unfinished work, today · built from P02, P05 and the survey

BetterWorse

relayDefine
1019

08 · Three roles in every handoff

The Inheritor

P02, P05 · engineers · 4.8 h/week
"I don't need their entire project history."

Needs What's done, blocked and settled; what could break; who to ask.

The Passer

P01, P04 · designers · 3.9 h/week
"I end up writing long Slack posts explaining what changed."

Needs A quick structure for decisions, and a sign the handoff landed.

The Coordinator

P03, P06 · PMs · 5.2 h/week
"A handoff isn't complete just because somebody posted a link."

Needs Opened, clarifying and accepted states; decisions recorded once.

relayDefine
1119

09–11 · Jobs, problem, How might we

People who inherit unfinished product work need to know what's settled, what's still a guess and what to do next, because the reasons stay in meetings and threads.
  1. T1How might we keep the reason next to each decision without adding paperwork?
  2. T2How might we make "decided", "assumed" and "left out" obvious?
  3. T2How might we keep evidence strength and limits attached?
  4. T3How might we show a receiver where to start in ten seconds?
  5. T3How might we count a handoff complete only when the receiver can act?
  6. T4How might we pull context from Slack, Figma and Linear without copying it?
relayIdeate
1219

12–13 · Opportunity map and prioritisation

Ranked by evidence,
not by excitement.

FeatureEvidenceImpactEffortPriority
Decision with a Why70.8% · P01, P03HighLowMust
Decision states incl. Out of scope60.4% · P02HighLowMust
Where it stands, next actions25.0% · P05HighLowMust
Open questions, Accept stepP02, P06HighMedMust
Sources that link out43.8% · P01, P05MedLowShould
Evidence strength and limitsP04MedLowShould
Auto-import from Slack, Linear43.8%HighHighLater
Task management, file storageBoundary–HighWon't
relayIdeate
1319

14 · Concept development

Three ways to carry the why.

Concept A

A handoff template

A structured canvas or page teams fill in.

  • + Cheap, uses existing tools
  • − No states: can't show settled or accepted
  • − One more document to find
Concept B

Annotations in place

A Figma and Linear plugin that pins decisions to frames.

  • + Context on the artifact
  • − Still split across tools
  • − No moment where work changes hands
Concept C · chosen

The handoff as an object

A standalone record of the transfer with an Accept step.

  • + Answers all four themes
  • + Sent, opened, accepted
  • + Links to tools, never copies
relayDesign
1419

15–16 · User flow, site map, wireframes

User flow
Fig. 1 Flow 1 of 4: inherit and accept. Open questions stay attached after acceptance.

Wireframes · layout before look

Incoming wireframe
Fig. 3 Incoming
New handoff wireframe
Fig. 4 New handoff
Handoff detail wireframe
Fig. 5 Handoff detail
Site map
Fig. 2 Site map. Two homes, Incoming and Sent, around one object.
relayDesign
1519

17 · A working prototype, not a click-through

New handoff · live preview, recording of the prototypeNew handoff · live preview
Write for the reader while you write. Four fields required; the preview shows what the receiver will see.
Ask about one decision, recording of the prototypeAsk about one decision
Questions attach to the decision they're about.
Accept, recording of the prototypeAccept
Accepting names what's still open and what could break.
Try it live · relaycontinuity.netlify.app
relayTest
1619

18 · Testing round one · heuristic evaluation and a cognitive walkthrough of four tasks

Seven issues.
Seven fixes.

7issues still open
1 Critical2 High2 Medium2 Low
Accept confirmation, before
Before
Accept confirmation, after
After · names the risks
CriticalError preventionA handoff with no risks crashed the detail screenIssue 01
FixedError preventionSample data is normalised on loadIssue 01
HighVisibility of statusAn empty "What could break" was hiddenIssue 02
FixedVisibility of statusAlways shown; it says when nothing was listedIssue 02
HighVisibility of statusAccepting didn't mention risksIssue 03
FixedVisibility of statusThe confirmation names what could breakIssue 03
MediumConsistency"Why" label on an Out of scope decisionIssue 04
FixedConsistencyThe label follows the type: "Why not"Issue 04
MediumHelp and documentation"How strong is it?" wasn't explainedIssue 05
FixedHelp and documentationA one-line definition under the fieldIssue 05
LowRecognition over recallContext check didn't count risksIssue 06
FixedRecognition over recallRisks are counted in the summaryIssue 06
LowAesthetic and minimalFields stretched to a taller neighbourIssue 07
FixedAesthetic and minimalFields align to the top of their rowIssue 07
relayTest
1719

19 · Testing round two · design critique, screen by screen

Every change removed something.

  1. 01Up next lost half its content

    Status paragraph, four tiles and a list went; title, sender, settledness and two actions stayed.

  2. 02One bar style everywhere

    Track, meter, timeline and journey share one glass tube.

  3. 03One left edge

    Chips, heading, button and ask box line up.

  4. 04Less chrome

    One "Demo user" pill; no banner; no chapter icons.

  5. 05Phones designed, not squeezed

    The track stands upright; chips become a three-part switch.

Hover a screen to enlarge it

01First design · bento
First design · bento
02Clay 3D
Clay 3D
03Glass round · final
Glass round · final
relayDeliver
1819

20 · Final product design

Your work, as a track.

Four stations show where each handoff is: just arrived, reading, questions out, yours. The urgent one is in coral.

Evidence · P05, P06

A handoff, read in chapters.

Sent, Opened, Questions, Accepted along the top; six chapters to jump between; the Transfer card with Accept stays pinned.

Evidence · P01, P06

Every decision carries its why.

Confirmed, Assumption, Proposed or Out of scope, with evidence strength and what it doesn't prove.

Evidence · P01–P04

The same track, from the other side.

Delivered, Opened, Questions in, Accepted. Senders see whether their handoff was even opened.

Evidence · P01, P06
Incoming, desktopHandoff, desktopDecisions, desktopSent, desktop
Incoming, phoneHandoff, phoneDecisions, phoneSent, phone
relayDeliver
1919

What I learned

Surveying and interviewing both sides changed who I designed for. A handoff looks like something the sender writes. The real effort is the receiver's.

NextRound 2 user sessions

5–6 people re-creating their own recent handoff; task success, SEQ, SUS.

QuestionIntegrations

Which connections pull context in without notification noise?

NextMeasure the baseline

Rerun the survey: 4.6 h a week, 18.8% find the rationale.

Back toAll work