Versions

Current release v1.4.0. Every version lists what was added, improved and fixed — and exactly what is mocked in it.

v1.4.0

current2026-08-28

A designed product, not a styled prototype

✨ Features

New landing page: the pitch told visually — a prism splits the thin official record into the five things that decide a purchase, and eight portals collapse into one on scroll
A prism wordmark: one beam in, a spectrum out — the same idea as the product
A real light/dark theme toggle; the app previously followed the operating system with no way to override it
Editorial display typeface (Eczar) covering both Latin and Devanagari, so Hindi headlines have the same voice as English

📈 Improvements

One accent colour, used once per screen on the action that matters. Its hue flips between themes so both grounds stay legible
Every page moved onto a shared frame; they had each invented their own width, padding and heading size, so the layout shifted as you moved between them
The disclaimer is now a considered line with a link to the evidence, instead of a black warning bar that made the product look unfinished
Sahayak rebuilt: step-aware subtitle, tappable prompts, typing indicator, auto-scroll, and the model picker demoted out of the conversation
Control borders raised to a 3:1 contrast ratio, which the previous palette failed

🔧 Fixes

Entrance animations hid content and relied on JavaScript to bring it back, leaving pages blank when a link was opened in a background tab

🎭 Mocked / synthetic (honesty ledger)

All data unchanged from v1.2.0 — see below

v1.3.0

2026-08-28

Design system, admin sidebar, and SEO

✨ Features

Design system: semantic success/warning/danger/info tokens that work in light and dark
Shared state vocabulary — skeletons, empty states, error states, spinners — plus route-level loading, error and not-found boundaries
Sidebar-first admin with an Overview dashboard and Owners, Challans and Applications screens alongside Vehicles
SVG wordmark replacing the emoji logo, and Lucide icons throughout
Full SEO: per-page metadata, OpenGraph and Twitter cards, generated OG image, sitemap, robots and JSON-LD
Prettier, Husky and a pre-commit hook that formats and lints staged files

📈 Improvements

Hindi finally renders in a real typeface: the previous font shipped no Devanagari subset, so every Hindi string had been falling back to whatever the operating system had
The vehicle search now shows a skeleton while loading and tells 'no such vehicle' apart from 'could not reach the record', with a retry
GSAP eases the Trust Report in, respecting prefers-reduced-motion; every other screen stays on CSS so the bundle stays small for slow connections

🔧 Fixes

A signed-out visit to the admin area redirect-looped between the guard and the login page
Admin pages streamed data to anonymous callers: a layout-only guard does not stop a page rendering, so the vehicle table was reachable without signing in

🎭 Mocked / synthetic (honesty ledger)

All data unchanged from v1.2.0 — see below

v1.2.0

2026-08-28

Database, real admin auth, and a data admin panel

✨ Features

Postgres (Supabase) backs the vehicle record: vehicles, owners and challans. Applications and payments still live in the browser — see How it works
Row-level security: the vehicle record is publicly readable, but only an admin role can write it
Admin panel at /admin — sign in, then create, edit and delete demo vehicles
Role model with super_admin / admin / viewer; signing in alone grants nothing
AI model picker: three free models plus paid gpt-oss-20b, defaulting to free

📈 Improvements

The app falls back to the built-in synthetic fleet whenever the database is unset, asleep or unreachable, so the citizen demo cannot go down with it
The admin panel states which source it is reading, mock or database
AI runs on OpenRouter with a server-side model allowlist, so a caller cannot name an expensive model and spend account credit

🔧 Fixes

Raised the AI token ceiling: several free models are reasoning models that returned empty answers when they ran out of tokens mid-thought

🎭 Mocked / synthetic (honesty ledger)

Every row in the database is synthetic. No real registration numbers, owners or phone numbers
Citizen login is still a demo OTP — real phone auth would mean handling real personal data
Admin accounts are real Supabase Auth accounts, created by us, not by citizens
All other mocks unchanged from v1.0.0 — see below

v1.1.0

2026-08-28

AI help, full Hindi, and an honesty page

✨ Features

Sahayak: a helper that knows which step you are on and answers paperwork questions in Hindi or English
AI-written Trust Report verdict via an OpenAI model, with a deterministic rule engine as fallback
Knowledge base of 8 bilingual answers (Forms 29/30/35, NOC, challans, PUC, fitness, accident rights) drawn from primary-source research
'How it works' page: who has the problem, what we changed, what is real vs simulated, and how it could scale safely
Keyboard skip-link to main content

📈 Improvements

Hindi now covers the whole demo path, not just the headings (dictionary grew from 26 to 70 keys)
Every AI answer is labelled with which brain produced it, so nothing overstates what is running
End-to-end test uses stable selectors, so improving the wording no longer breaks CI

🎭 Mocked / synthetic (honesty ledger)

Sahayak and the verdict use the offline rule engine unless OPENAI_API_KEY is set; the screen says which one answered
All other mocks unchanged from v1.0.0 — see below

v1.0.1

2026-08-28

Post-review hardening

📈 Improvements

One shared login gate across garage, report and transfer instead of three copies
Single source for the demo date and currency formatting, so no two screens can disagree
Applications can now be advanced to COMPLETE (Simulate RTO approval in My Garage)
Removed 11 unused dependencies and 5 unused components the code never imported
CI now runs the end-to-end smoke test, not just lint and build

🔧 Fixes

Design tokens were missing, so every button and card rendered unstyled — the whole UI was flat text
Reloading while logged in caused a React hydration error and a flash of the logged-out screen
Application numbers were 5 digits while the status page asked for 6, so valid lookups failed
Paying then abandoning the consent step charged you with nothing unlocked, and charged again on retry
Submitted transfers were stuck on 'IN PROGRESS' forever and could never complete
A negative loan tenure produced a nonsense EMI presented as real financial guidance
The page language attribute stayed 'en' in Hindi mode, mis-cueing screen readers
Footer floated mid-screen on short pages

🎭 Mocked / synthetic (honesty ledger)

Unchanged from v1.0.0 — see below

v1.0.0

2026-08-28

Hackathon v1 — the used-vehicle citizen journey

✨ Features

One citizen account: single mobile + OTP login shared across every service (demo OTP)
Vehicle check: free summary card mirroring today's official masked lookup (~10 fields)
Trust Report: consent-unlocked full report — ownership timeline, loan/hypothecation panel with financier, challan history, insurance/PUC/tax/fitness validity, accident flag, fair-price band
EMI affordability calculator inside the loan panel
AI verdict: plain-language buy/don't-buy explanation of the report (English/Hindi)
Guided transfer of ownership: one wizard for Form 29/30 + HP termination (Form 35) + fee + e-sign + RTO slot, with a persistent stage tracker
My Garage: my vehicles, applications with live stages, payment history with receipts, expiry nudges (insurance/PUC/tax/fitness)
Crash Card: virtual doc view, 112 one-tap, nearest cashless-scheme hospital, Good Samaritan (Rahveer) rights
Application status lookup by application number
English/Hindi language toggle
In-app changelog and versioning (this page)

📈 Improvements

Replaces 8 disconnected portals, 4 staff-only logins, per-portal state gates, and captcha-on-every-lookup with one responsive web app

🎭 Mocked / synthetic (honesty ledger)

All vehicle, owner, challan, and accident data is synthetic (8-vehicle demo fleet)
OTP is a fixed demo code; no SMS is sent
Payments simulate a gateway; no money moves
Seller-consent unlock mirrors the MoRTH Data Sharing Policy consent model — proposed, not a live integration
AI verdict is rule-generated in the demo; OpenAI-powered when an API key is configured

Roadmap (walkable previews, not live services)

Scoped to Gujarat, because that is the state the research walked. Each one runs end to end on synthetic data under All services; none of them files anything with an RTO. The reasoning and the numbers behind them are on How it works.

Commercial lane: fitness and ATS booking, MV tax, national and state permits (Gujarat: 65 ATS centres, ₹193 Cr/yr in national permits)
Licence journeys: learner licence and DL renewal, on top of Gujarat's 28 contactless eKYC services, with the Form 1A empanelled-doctor finder
Fancy-number auctions you can browse before logging in (40 Gujarat RTO codes)
Scrapping: Certificate of Deposit into a discount on the next registration
One grievance ticket per failure, instead of four separate systems
OpenAI-powered verdict and Krishna chat on live keys; Supabase persistence behind the existing data interface