SPacial INteraction design • project 2024
Taking a farm loan from a bank branch queue to the field.
ROLE
UI Designer & Spatial interaction Designer
TIMELINE
August 2024 - October 2024
TEAM
2 Designers
4 Developers
3D Environment team
SKILLS
UI Deisgn
VR Design
Prototyping
Challenge
Can a farm loan happen in a field, not at a bank branch?
The Kisan Credit Card (KCC) gives Indian farmers access to crop credit. Getting one does not match the life of the person it serves. Applications live across fragmented bank portals, the process stays largely manual, and a farmer typically travels to his nearest branch and waits.
RBI set out to replace that with a single digital flow, carried into the field on tablets by their own agents. We had to serve the following:
The application itself
A unified KCC journey from identity to disbursement, usable by an agent in a field and by a farmer on his own.
The training for it
RBI's agents needed to learn the flow before meeting a farmer.
Adapt the app to VR
The Government of India already had a metaverse environment in development, so the training would happen in VR, with agents rehearsing full scenarios in a headset: the conversation, the farmer, and the interface, all at once.
The Two Users
The product serves farmers.
The product is operated by agents.
RBI made a deliberate call to allocate research time to the agents, since they are the ones using the tablet day after day.
The agent
fills his fifteenth form of the day. He needs speed, predictability, and an interface he can operate without looking down, because looking down means looking away from the farmer.
The farmer
is doing this once, with money and land records on the line. He needs reassurance, plain language, and a sense of what happens next.
Everything downstream of the first screen had to serve both.
Research
We combined secondary research on the existing KCC process with conversations with three RBI field agents. Those conversations shaped the flow more than anything else.
Insight 1: The form is not the hard part, the waiting is
Agents described the awkward stretches where the system processes and nobody speaks.
Action
We designed every loading, scanning, and processing state as a moment the agent can talk through, and later tied those states to dialogue beats in the VR training.
Insight 2: Trust lives in the thumb
Agents told us the biometric scan is the moment farmers understand something official is happening.
Action
We gave the fingerprint flow its own full-screen treatment with explicit states, and reused it at both identity verification and e-signing.
CorE FLow
The original process asks a farmer to assemble documents, travel, queue, and submit. We rebuilt it around a single question:
What does the farmer actually need to supply?
Action
We reduced farmer-supplied input to two values and fetched everything else from government systems.
What this bought us
The two heaviest verification steps in the entire process cost the farmer twelve digits, three digits, and a thumb. Fetched data also removes transcription error, which is a common cause of rejected applications.
What it cost us
The flow now depends on two external systems being available and accurate. Land records in India are patchy, and a survey number (connected to Tax ID) that returns nothing or returns the wrong owner needs a recovery path. That path lives outside the current flow, and building it into the product is the clearest next step.
Design tensions
Two users, one flow
Tension: A self-serving farmer and a working agent need opposite things from the same screens.
Decision: I split the paths at the first screen, Self Assisted and Bank assisted, and routed the agent through an Employee Login before the farmer’s mobile number. Everything after that is shared.
Verified records versus the farmer's own knowledge
Tension: Prefilled government data speeds the flow and keeps the record legally clean. It also presents those records as settled fact, and the farmer may know his address has changed.
Decision: Fetched fields stay read-only. I added one editable path: a separate communication address with a “Same as Permanent Address” toggle.
Structured data versus how farming actually works
Tension: The credit model needs crop data. Indian smallholders rarely grow one crop on one clean plot, and 0.7 hectares might carry both sugarcane and rice.
Decision: A repeating crop block with “Add More Crops,” anchored to a persistent “Total Declared Land Area” so the constraint stays visible while the farmer allocates within it.
The sanctioned ceiling versus the right amount
Tension: Showing a sanctioned maximum of ₹80,000 maximises disbursement. Borrowing the maximum is rarely in a smallholder’s interest.
Decision: A slider between a minimum and the sanctioned ceiling, with processing fee, interest rate, and tenure recalculating live as the farmer moves it. PMFBY crop insurance sits on the same screen as an opt-in.
Legal completeness versus comprehension
Tension: A loan agreement must be presented in full and consented to in a legally valid way. A dense block of declaration text is not something most farmers in this segment will read.
Decision: I kept the full terms intact and worked on the consent around them. Two e-sign methods, OTP and biometric, so a farmer who cannot manage an OTP can still sign with his thumb. An Agreement Preview with a download option sits before the final commit.
CHanges to the design Systems
An RBI design system already existed across other products, built for urban, literate, banking users. It had a palette, a component set, and no answers for the problems this product raised. My work was extension and codification, not creation.
Bilingual label pairs
Self Assisted / स्वयं करे sits as one two-line unit with fixed hierarchy between the scripts. Defining it as a single component stopped every bilingual moment from inventing its own treatment.
Biometric hardware states
The fingerprint flow has four states: device not connected, connected, scanning, and successful. I defined it as one component with those states rather than as four screens. That let it reappear unchanged at e-signing, which is exactly where the flow needed it a second time.
Taking the Screens into VR
The 3D team had built the environment: a village, a farmer, an agent, and scripted scenarios.
My job was making the interface work inside it.
Positioned the panel inside the conversation
The panel sits between agent and farmer, close enough to read and far enough to keep the farmer’s face in view. The training exists to teach that the farmer matters more than the screen, so the layout had to make that possible.
Synced screen states to dialogue beats
The scenarios came with scripted dialogue. My screens came with their own rhythm: OTP waits, fingerprint scans, processing states. I aligned the two, so dead time in the interface became the moment the agent speaks. The fingerprint scan is when he explains Aadhaar eKYC. The processing screen is when he reassures.
Built a haptic hierarchy scaled to interaction weight
A floating panel has no surface, so a press can feel like nothing happened. Feedback loud enough to pull the agent’s eyes back to the panel is the opposite failure. Working in Meta’s haptic design document format, I mapped every event by type, frequency and priority, then prototyped four levels:
IMpact
Piloted, presented to the RBI Governor, and now used in agent training
By design, a farmer’s input dropped to two numbers and a thumb, with every other identity and land detail fetched from government records. Two reusable components, bilingual labels and the biometric flow, gave the system one consistent treatment wherever it appeared. And because the same screens power both the field app and the VR module, agents train on exactly what they’ll use with farmers.
.
saraSanwal2026™