1Overview
India has built the rails of a digital state — identity, payments, documents and language technology at a scale no other country has attempted. What it doesn't yet have is a front door that feels made for the person standing in front of it.
Saathi reorganises government services around what happens in a citizen's life, speaks their language, and adds an assistant that can prepare an application, show its plan, ask for consent and hand back proof.
What was designed
42
government processes modelled on their real portals
17
languages today — the other 6 of the Eighth Schedule are coming soon
10
onboarding steps, including consent under the DPDP Act
1
assistant that never submits without your yes
2The problem
A citizen who wants to register a birth, claim a pension, or check a land record doesn't deal with one government. They deal with dozens — each with its own app, its own login, its own language, and its own idea of how a form should work.
Services exist, but finding the real one — not a lookalike, not a middleman — takes confidence, data, and patience that the people who need help most often don't have. The maze itself is the barrier.
Saathi asks a simpler question: what if there were just one door?
A few of the apps a citizen is expected to juggle
मुझे पता है कि सब online है, but मुझे ये नहीं पता कि कौन सा online असली है.I know everything's online — I just don't know which 'online' is the real one.
3Saathi and UMANG
UMANG is the government's own unified app — thousands of services in one place. It proved aggregation is possible. Saathi starts where UMANG stops: not 'how many services fit', but 'how does it feel for a real person to use one'.
| UMANG | Saathi | |
|---|---|---|
| Organising idea | A directory of departments and services — you must already know the exact name. | Life moments, needs and entitlements — you start from your situation, not a department. |
| Language | 13 languages, but most forms still fall back to English. | Vernacular-first, voice on every field, and the whole UI swaps script in 200ms. |
| Discovery | You search — nothing tells you what you're actually owed. | Profile-based scheme cards with a transparent 'why you qualify'. |
| Trust & calm | Official, but visually dense and utilitarian. | Calm authority — verification and status surfaced in the first five seconds. |
| Help & status | Largely on your own once you submit. | A status timeline with named officers, and family-proxy as a first-class action. |
| Accessibility | Standard forms; little for low-literacy or low-vision users. | Slow mode, larger text, read-aloud and offline-first — on by default. |
UMANG is a real, large-scale product serving millions. This comparison is about design philosophy — not a critique of its engineering or reach.
4Research
Over several weeks I spoke with people in cities, small towns and villages about the last time they dealt with a government service. The same three frictions surfaced again and again: language, trust, and simply not knowing where to start.
- R
Rakesh Tier-2 city
Has a smartphone and data, but can't tell the real portal from a lookalike.
- S
Sunita Rural
Relies on a CSC operator for everything — and pays for it.
- V
Vikram Senior
Wants it read aloud, in Tamil, without rushing.
- A
Ananya Urban
Fast, fluent, English — wants it done on the train.
5Personas
Rakesh Verma
34 · Tier-2 · Kanpur
Shopkeeper. Smartphone-confident, but burned by fake-portal scams.
Wants a birth certificate for his newborn without taking a day off or paying an agent.
“If it looks official, I still double-check. Twice.”
- Trust signals he can verify
- Hindi-first interface
- Clear status, no jargon
Sunita Devi
29 · Rural · Bihar
Farmer. Low literacy, shares one phone with the family.
Wants to claim a maternity benefit she heard about but can't read the form.
“I can speak it. I can't always read it.”
- Voice input and read-aloud
- Large taps, few steps
- Works on a weak signal
Vikram Singh
68 · Senior · Chennai
Retired teacher. Patient, careful, prefers Tamil.
Wants his pension status without calling his son every time.
“Don't hurry me. Just tell me clearly.”
- Larger text and slow mode
- Audio labels
- Tamil throughout
Ananya Sharma
26 · Urban · Bengaluru
Analyst. Fluent in English, zero patience for friction.
Wants to renew a document on her commute, in two minutes.
“If it takes more than the train ride, I'll abandon it.”
- Speed and keyboard flow
- Receipts she can save
- No hand-holding
How each person uses Saathi
| Persona | Entry point | Key need | Voice |
|---|---|---|---|
| Rakesh | Moments | Verifiable trust | Sometimes |
| Sunita | Voice | Read-aloud forms | Always |
| Vikram | Entitlements | Slow, clear, Tamil | Often |
| Ananya | Search | Speed | Rarely |
6The frame
Every service in Saathi is reachable through three lenses that meet at one person.
- Moments
- Life events — a birth, a job, a move, a loss.
- Needs
- What to do right now — a document, a payment, a proof.
- Entitlements
- What you're owed — schemes, pensions, subsidies.
7Principles
- 1
Lives, not departments
Organise around life moments, not government silos.
- 2
Vernacular-first
Designed in Hindi and Tamil first, then ported to English.
- 3
Trust in 5 seconds
Verification, .gov.in signals and calm authority, up front.
- 4
Status is sacred
Always show where a thing is, who has it, and by when.
- 5
Proxy is normal
Helping a parent or child is a first-class action, not a hack.
- 6
Voice is first-class
Every field can be spoken, not typed.
- 7
Receipts everywhere
Every action ends with proof you can keep.
- 8
Density is the enemy
Few choices per screen. Space is a feature.
- 9
Inclusive by default
Built for the lowest-literacy, lowest-bandwidth user first.
- 10
Calm authority
Official, never cold. Reassuring, never childish.
8Structure
Version one had an Apply tab — but a bare Apply tab asks “apply for what?”. The centre of Saathi 2.0 is the assistant: tell it what you need, in any words, and it routes you. Every one of the 42 flows is still reachable from Explore, detail screens and Home.
Home
Identity, what needs attention, your applications and schemes for you.
Explore
Life moments, services and schemes — plus search in any language.
Saathi
The assistant. Ask, see the plan, approve, get proof.
Track
Every application with stages, named officers and dates.
Me
Document wallet, family help, privacy dashboard and preferences.
9Designing Saathi AI
Most government chatbots answer questions. The people we designed for don't need more answers; they need the thing done. So Saathi AI is designed as an agent: it plans, uses official services and produces an outcome.
Acting on someone's behalf with their identity and money raises the bar for trust. Every pattern below exists to keep the citizen in control.
Four patterns
1Plan card
Before doing anything, Saathi lists its steps and the official service behind each one. The plan updates live as steps complete.
2Tool steps
Each action — fetching a record from DigiLocker, checking eligibility — appears as a short, verifiable line with its source.
3Consent gate
Nothing is submitted, paid or shared until the citizen approves a card that names the data, the recipient and how long access lasts.
4Outcome
Every mission ends in something real: an application in Track, a paid receipt, a switched language — with a reference number.
When it can't help
- Didn't understand
- Offer three likely requests in plain words, and the option to speak to a person.
- Not eligible
- Say so plainly, explain which rule applies, and suggest what the person may qualify for instead.
- Service unavailable
- Keep the prepared application as a draft and say when to retry. Never lose what someone has entered.
- Needs a human
- Hand over to the nearest Common Service Centre or a helpline, with the context already filled in.
AI principles
- Official sources only, always named
- Asks before it acts on identity, money or data
- Shows its reasoning for every eligibility claim
- A person is always one tap away
- Scripted and reviewable — no hidden prompts deciding outcomes
In this prototype the assistant is pre-scripted, not a live language model — so every answer it gives can be reviewed before it reaches a citizen.
10Languages
Three levels of going local
1
Translation
The same words, rendered in a new script.
2
Localization
Local norms too — dates, currency, names, number formats.
3
Transcreation
Rebuilt for the language. Designed in Hindi first, then ported.
- Per-script line-height keeps Devanagari and Tamil legible.
- Names render in the reader's own script.
- The whole UI swaps in 200ms — no reload.
Switch the language of this page
Every screen, the assistant, the department consoles and this case study are available in 17 languages today. Hindi, Bengali and Tamil were written first, alongside English; 13 more are machine-assisted and marked for native-speaker review. The other 6 languages of the Eighth Schedule are listed as coming soon.
11Accessibility
Voice on every field
Speak any answer instead of typing it.
Slow mode
More time, fewer steps, gentler pacing.
Larger text
Scales up without breaking the layout.
Audio labels
Buttons and labels read aloud.
Family / proxy mode
Act for a parent or child, clearly framed.
Offline support
Designed for weak and intermittent signal.
Colour-blind safe
Status never relies on colour alone.
Screen-reader ready
Semantic structure and live announcements.
Every text colour, tested
Contrast ratios are computed from the design tokens in code and checked in automated tests against WCAG 2.1 AA.
| Pairing | Ratio | Result |
|---|---|---|
| Aaink on base | 19.41:1 | AA |
| Aaink-2 on base | 10.54:1 | AA |
| Aaink-3 on base | 5.51:1 | AA |
| Aaink-3 on surface | 5.19:1 | AA |
| Aaink-2 on surface-2 | 9.24:1 | AA |
| Aanavy-600 on base | 8.22:1 | AA |
| Aanavy-700 on navy-50 | 9.84:1 | AA |
| Aawhite on navy-600 | 8.22:1 | AA |
| Aawhite on navy-900 | 16.07:1 | AA |
| Aasaffron-700 on base | 5.02:1 | AA |
| Aasaffron-700 on saffron-50 | 4.70:1 | AA |
| Aaink on saffron-500 | 8.23:1 | AA |
12Design system
A design system calm enough to trust with a pension and precise enough to feel world-class — on a pure white base, with colour reserved for meaning.
Colour with a job
Chakra Navy
#1F43B8
Actions, links, focus
India Green
#12803C
Verified, approved, done
Saffron
#FF8A1F
Attention and highlights, never body text
Ink
#0B0D14
Text
White
#FFFFFF
Every surface
Aa अ த
Type across scripts
Geist for Latin, Anek for Indic headlines (by Ek Type, an Indian foundry) and Noto Sans for Indic body text at a 1.8 line height, as UX4G recommends.
Imagery made in code
No stock photos and no maps. Kolam linework for texture, guilloche security print for anything official.
Motion with a reason
Interface motion stays under 300 milliseconds and can always be interrupted. Storytelling motion breathes. Everything becomes a gentle fade for people who prefer reduced motion.
Saathi is a speculative civic project with its own identity — two arcs and a point, drawn from the tricolour — paired with a clear design-concept disclaimer. It never uses the State Emblem of India, which is legally protected.
13Key flows
Onboarding
From install to ready, in your language.
Language, Aadhaar, OTP, a quick profile and accessibility setup — nine calm screens.
Open this flowLife Moments
Find services by what's happening to you.
A new baby surfaces birth registration, vaccination and benefits — together.
Open this flowVoice Form Entry
Fill a form by speaking.
Tap the mic, speak, confirm the structured answer, and the field fills itself.
Open this flowTrack Application
Know exactly where things stand.
A vertical timeline with completed, current and pending steps — and named officers.
Open this flowWhat I'm Owed
Discover schemes you qualify for.
Profile-based scheme cards with transparent 'why you qualify' reasons.
Open this flowFamily Proxy
Do it for someone you care for.
A persistent ‘helping Sunita’ banner keeps every action in the right context.
Open this flow
14Measuring success
These are design targets for a pilot, not results. They are here so the design can be judged against something measurable.
Design targets
5 min
to apply for a birth certificate, from a first-time user's first tap
90%
of test participants complete a key flow without help
0
applications submitted or payments made without explicit consent
1 in 3
form fields filled by voice among low-literacy participants
Validation plan
- 1
Usability testing in three states
Moderated sessions with each persona group, in their language, on their own phones.
- 2
Native-language review
Every language reviewed by native speakers before any pilot — the 13 machine-assisted ones first.
- 3
Independent accessibility audit
Screen reader, keyboard, low vision and cognitive load, against WCAG 2.1 AA and GIGW 3.0.
- 4
Assistant red-teaming
Try to make Saathi AI submit, pay or share without consent — and fix every path found.
15Reflection
Scoped out for now: real Aadhaar and payments, live speech recognition in every language, and an offline-first data layer.
Next, I'd pressure-test the voice flows with low-literacy users, put every language in front of native reviewers — the 13 machine-assisted ones first — and bring in the 6 still to come.
About Krishom
I'm a product designer working on civic and consumer products. Saathi is a self-initiated exploration of what dignified, inclusive government software could feel like.
Get in touchOther case studies