
Why this project?
Every winter in Delhi, people sleeping rough go unnoticed by commuters who would help if they knew how, and felt safe doing it.
In early research I saw the same pattern again and again. Someone notices a person in need and wants to help. They hesitate because they don't know who to call or whether it's safe. Within seconds, they walk on. AIDE is built to catch that moment.
17.7 lakh
Houseless people counted in India (1 lakh = 100,000)
Census of India, 2011
1 in 3
Said a government partnership would make them trust a helping app
AIDE survey, n = 34
1 in 5
Said lack of time stops them from helping
AIDE survey, n = 34
Project Snapshot
Constraints
Eight constraints shaped every screen.
No centralized NGO database
NGOs had to be discovered and contacted individually.
Privacy-first reporting
Personal information could not be mandatory to report someone.
Need anonymous reporting
Citizens hesitated to help if their identity was exposed.
Cold-weather urgency
Reports needed to reach responders fast, with honest wait times.
No beneficiary smartphones
Homeless individuals could not interact with the application themselves.
Volunteer uncertainty
Volunteer availability changed daily and couldn't be guaranteed.
No government APIs
Shelter capacity and official data couldn't be fetched automatically.
Low digital literacy
Reporting had to work with minimal instructions and few steps.
Research
I wanted to understand not just who needed help, but what stopped people from offering it.
34 survey responses pointed to the most common barrier: not knowing how to help.
Core Survey Insights
67%
Empathy is the dominant first response
Two in three named empathy as their first response.
44%
The biggest barrier is not knowing how
Said not knowing how to help effectively holds them back.
74%
Verified impact drives repeat action
Would help more if they knew their actions made a real difference.
What this means
The design problem is not motivation. It is the missing step between feeling and acting, and the proof that acting mattered.
FIndings
I surveyed 34 people across different age groups in Delhi to understand what stops them from helping, what builds trust, and how much time they're willing to spend taking action.
Trust
Distrust of the process
People don't act due to lack of time, clarity, or confidence that help will arrive.
Triggers
Motivation spikes, then vanishes
Cold weather, children, visible suffering: emotional triggers create a brief window. If nothing catches it, it closes permanently.
Approach
Indirect help first
People strongly prefer to report, donate, or share. Safety is the non-negotiable condition for any action at all.
Closure
Knowing it worked is the real reward
Confirmation that help arrived is a stronger motivator than badges, points, or any reward system designed to date.
Interview
17 informal interviews traced hesitation to trust and effort.
I followed up with one-on-one interviews to understand the stories behind the survey results. Speaking with citizens, NGO workers, and people who had experienced homelessness helped me uncover the emotions, doubts, and motivations behind every decision.
Group A
General Public
Commuters
8 participants
"It's not that I don't want to donate. I just need to know it's going to the right place."
Anjali, 19, Student
Group B
Recipients
Living rough
7 Participants
"Most people avoid eye contact. It makes you feel invisible, and that's the hardest part."
Laxmi, Participant
Group C
NGO Staff
Unofficial Volunteers
2 Participants
"The hardest part isn't always finding people to help. It's knowing where help is needed and getting there in time."
Monisha, 27, Volunteer
Interview insight
Everyone got stuck at the same point, unsure what to do next or where to turn for help.
Most people wanted to help. By the time they figured out what to do, the moment had often passed.
I synthesized the findings into two primary personas, one on each side of the experience.
General Public
Beneficiary
Users
Two people, separated by everything except one cold night.
My research revealed two very different perspectives on the same moment on the street.
General Public
One wanted to help.
Urban commuters who notice distress and feel the urge to act, then stall at the question of who to call and whether it is safe.
Beneficiary
The other simply wanted to be seen.
People living rough, usually without a phone, for whom being noticed is the first form of help and avoided eye contact is the hardest part.
User Journey Map, General Public
One reporter's moment, from noticing someone to deciding what to do.

After research
Problem statement, redefined.
From “help the homeless” to “remove the friction between wanting to help and helping.”
Intent
People want to help.
People often want to help individuals in visible distress.
Decay
Intent fades fast.
As hesitation grows, empathy turns into avoidance, delayed intention, or ineffective help.
Uncertainty
Time pressure
Safety concerns
No trusted channel
Redefined
Reduce the friction between empathy and action.
AIDE gives citizens a way to connect vulnerable individuals with verified support systems that is:
Fast
Trustworthy
Low-commitment
Solution
The Empathy Decay Problem
The urge to help fades with every second of uncertainty.

Conceptual illustration
People often want to help, but uncertainty and effort delay action. As time passes, the likelihood of acting decreases and the opportunity to help is lost. AIDE reduces the effort required to act while intent is still high.
How research drove design decisions
Five research insights, five design decisions.
Behavioural Design Strategy
The centre is the first two seconds. The outer ring is who you become.
Each ring removes one psychological barrier I found in research. Together they turn a passing moment of empathy into a habit.
- Emotional CaptureReport first, details after, in a few steps. Lower friction than walking away. Replaces the instant gratification of a cash reward.
- Trust ArchitectureNGO verification. Anonymous default. A named no-police policy.
- Emotional ClosureStatus updates until the case closes, ending with what actually happened, even when no one could help. The open loop closes.
- Social BelongingCity-wide counts of people helped and anonymous success stories. You are part of something.
- Civic IdentityWinter Hero. Trusted Citizen. Titles you can call yourself, not points.
Key Design Decision
Why didn't I use cash rewards?
Rejected
Cash rewards
Cash incentives can shift helping from compassion to compensation. They are difficult to sustain, attract the wrong incentives, and don't fit the realities of resource-constrained NGOs.
My approach
A habit loop
Rather than paying people to help, I designed a habit loop that brings people back because helping feels meaningful. Every interaction reinforces:
Trust
Visible impact
Belonging
Recognition
Product impact
Helping becomes a habit. The experience is sustainable for NGOs and rewarding for citizens without relying on financial incentives.
Inspired by principles from Hooked by Nir Eyal
AIDE Sitemap
Designing for urgency and intention.

The same person interacts with AIDE in different emotional states.
Sometimes they need to act immediately before empathy fades.
Other times they seek meaning, community, and evidence of impact.
The experience adapts to both, balancing rapid intervention with long-term engagement.
End-to-End Reporting Flow
Every report reaches an outcome, even when no one can help.
I mapped every possible path a report could take, including emergency cases, anonymous reporting, duplicate detection, NGO assignment, and status updates. This ensured the experience remained simple for citizens while supporting the operational needs of NGOs.

Report in Seconds
The reporting flow is intentionally short. Users can submit a report with minimal information while optional details are collected only when relevant.
Smart Processing
Emergency cases, duplicate reports, anonymity preferences, and NGO availability are handled automatically behind the scenes.
Close the Loop
Reporters get status updates until the case closes, including when no one could help.
What NGOs get
Duplicate reports merge into one case, so teams don't send two people to the same spot. Auto-location cuts the time spent finding the person. Every case has an accept step, so someone owns it from the first minute. These are design intents I'll measure with NGO partners.
Trust Architecture
Research showed trust builds through small reassurances at each step, from sending a report to case closure.

Civilian Support
Only
Clear communication that reports are routed to NGOs, not the police, addresses one of the biggest concerns uncovered during research.
All 5 participants drew the line at police. Reports go only to verified NGOs; calling 112 stays the reporter's choice.
Rejected: alerting the police automatically. I considered it to make the app feel safer, as many reporting apps do, but it breaks the one line all five participants drew.
Privacy First
Anonymous reporting removes the fear of personal exposure before users even begin.
Verified Every Step
Government affiliations and verified NGO badges build confidence before a report is assigned.
Transparent
Follow-up
Status updates and case closure reassure citizens that their report led to real action.
Wireframes
Screen Designs
Includes the main flow, 8 edge cases and 4 trust moments.
Play the prototype
Primary Platform
Mobile
Domain
Public Good
Context
Fieldwork
Scope
End to End




Testing the Experience with Users
Five people tested V1. All five stopped at the same screen.
I conducted moderated usability sessions with five participants to validate the reporting flow, identify friction, and verify whether the core trust mechanisms were discoverable without guidance. Two of the five, Brijesh and Monisha, had also taken part in my interviews. They knew the problem going in, which may have shaped how they read V1.
Findings
Six findings that shaped V2
Finding 01
5 of 5 participants
An inline 112 strip replaces the pop-up
All five paused at the “Critical Status Detected” pop-up pointing them to 112, India's emergency number. The woman in the scenario was awake, and nobody could tell whether to call, carry on or go back.
“Do I have to call first, or can I continue?”
Brijesh, 45
“I thought I'd selected something wrong or that the app was telling me to stop.”
Jatin, 19
In V2: the report opens with “Is anyone in danger right now?” If yes, an inline strip says “Call 112 now. We're still alerting NGOs nearby.”
V1 · Before

V2 · After

Finding 02
5 of 5 participants
Photo consent, and a privacy clause
All five hesitated at the photo step, and none wanted to photograph her face. V1 opened the camera straight away and made skipping feel like an excuse.
“I don't want the app to encourage people to photograph someone who can't consent.”
Smiti, 25
“I would prefer a skip option that doesn't make me feel I'm doing something wrong.”
Monisha, 27
In V2: the camera stays off until chosen, Skip and Take photo carry equal weight, a consent note sits above them, and faces are blurred before upload and deleted when the case closes. The cost: without a photo, responders have less to go on when they arrive. That trade-off is untested; it's first on my list to test with NGO staff.
V1 · Before

V2 · After

Finding 03
5 of 5 participants
An anonymity clause, and a phone number only if they want one
Asked who sees the report, everyone said “the NGO, probably”, then asked whether their own details went with it.
“The NGO, probably. But I want to know whether my name or phone number goes with it.”
Brijesh, 45
“I expect a notification when someone accepts it.”
Jatin, 19
In V2: “NGOs see the report, not who sent it.” After sending, an optional phone number for updates, never shown to NGOs, and a claim code for everyone else.
V1 · Before

V2 · After

Finding 04
3 of 5 participants
Short tips instead of a course
Nobody would sit through an 8-lesson course. The skill people wanted fits in a sentence: how to approach someone respectfully.
“You need to approach respectfully, understand the situation, and avoid making promises you can't keep.”
Smiti, 25
“I introduce myself, explain why I'm there, and let them decide whether they're comfortable talking to me.”
Shubham, 27
Cut: the 8-lesson course and its XP. Learn becomes 5-second tips from NGO outreach staff, one behaviour each. Recognition moves to badges, earned only after a confirmed outcome.
V1 · Before

V2 · After

Finding 05
Raised by Brijesh
Hindi and regional languages
Participants read English comfortably, but Brijesh pointed to those who can't: his parents' generation, and people seeking help for themselves.
“Even if they want to help someone, they might struggle to understand what each button means or what they're supposed to do next.”
Brijesh, 45
“Especially in an emergency, you shouldn't have to spend time translating things in your head.”
Brijesh, 45
In V2: an English | हिन्दी switch where seeking help starts. Next: regional languages and read-aloud instructions.
V1 · Before

V2 · After

Finding 06
2.8 / 5 trust
No police, and how the promise is kept
All five drew the same line: police for an emergency, never for someone who is homeless. Asked how far they trusted a report to stay anonymous and away from the police, participants rated V1 2.8 out of 5.
“Homelessness itself shouldn't automatically become a police matter.”
Monisha, 27
“In other cases, I would want to understand why the police need to be involved.”
Shubham, 27
In V2: the promise appears at three moments and the routing backs it. Reports go only to verified NGOs, and 112 is always the reporter's own call.
V1 · Before

V2 · After

The pattern
4 of 6 findings
Trust, not navigation, caused the friction
Four of the six findings are about trust: whether calling 112 cancels the report, whether a photo exposes someone, who sees the report, and whether the police get it. Smaller findings also shaped V2: optional fields for strangers, statuses that explain themselves, “help given” instead of “reached”, new screens for shelter and volunteering, and five edge-case screens.
Changes in V2
Paused at the 112 pop-up
V1 trust: anonymity and police
Session length
Reflection
Friction decided whether people acted.
The biggest lesson from 14 months of designing AIDE: people already wanted to help, so the work was making it effortless to act before the moment passed. V1 built the infrastructure. V2 was designed around human behavior.



