In a government office, work moves on paper and phone calls. A file sits on somebody's desk and nobody knows. A deadline passes and nobody notices until a citizen complains. When something goes wrong, there is no reliable way to find out who changed what, or when.
SMARTWORK 360 fixes three things:
| Problem today | What this does |
|---|---|
| Nobody knows where a file is stuck | Live dashboards for staff, managers and the collector |
| Deadlines slip quietly | Automatic SLA countdown, warnings before a deadline is missed |
| Records can be changed without a trace | Every change is sealed in a chain that shows if anyone edited it |
| Nobody notices when staff are drowning | The system spots overload and low morale from the work itself |
docker compose up -d # start the database (skip if postgres already runs)
npm install # install everything
npm run seed # fill it with a realistic district office
npm run dev # open http://localhost:3000Click a role on the login screen — no password typing needed.
Every demo account uses the password Demo@123.
| Sign in as | What you see | |
|---|---|---|
| 👑 Collector (Admin) | rajesh.iyer@gov.in |
Every department, fraud alerts, the audit chain |
| 👔 Manager | anil.kulkarni@gov.in |
One department's team, workload and morale |
| 👤 Employee | kavita.joshi@gov.in |
Only her own tasks |
Four pieces, in plain terms:
- Browser — what people see. Dashboards, task lists, charts. Switches between English and हिंदी instantly.
- API server — the rulebook. It decides who may see what, whether a task is allowed to move to the next stage, and it saves everything.
- Audit chain — a sealed copy of every change. Explained below.
- AI service — reads the notes people write and works out mood, overload and suspicious behaviour.
It keeps working with no internet. If the AI service is switched off, the API does the same calculations itself. Nothing on screen breaks or goes blank.
Every action — creating a task, approving it, adding a note — becomes a block. Each block carries a fingerprint made from the block before it.
So if somebody opens the database and edits one old record, its fingerprint no longer matches, and every block after it stops matching too. The change cannot be hidden.
Try to break it yourself:
npm run demo:tamper # secretly edits one record, straight in the databaseNow open Blockchain Audit in the app and press Verify chain. The screen turns red and names the exact block that was touched. The row still looks completely normal in the database — only recomputing the chain reveals it.
npm run demo:reset # put everything backMeasured: 962 blocks checked in 15 milliseconds.
Being straight with you: this is not a public blockchain and we never say it is. It is a sealed chain of fingerprints stored in an ordinary database, plus checkpoints every 100 blocks. Publishing those checkpoints to a real chain is designed but not built, and the screen says "planned" where that would go.
No chatbot pretending to be clever. Four specific jobs, each measured.
It scores the notes staff write — including Hinglish like "delay ho raha hai" — and shows the team's morale as a dial.
We tested a famous AI model against our own simpler method and the simple one won:
| Method | Accuracy on unseen notes |
|---|---|
| ✅ Our word-based method (shipped) | 87.5% |
| DistilBERT AI model, helped along | 85.0% |
| DistilBERT AI model on its own | 65.0% |
The AI model was trained on movie reviews and only knows "good" or "bad". Most lines in a government file are neither — "Placed the muster roll before the accounts branch" is just routine, and the model called it negative. So we shipped the method that actually works here, and we show the numbers rather than hide them.
From workload, missed deadlines, late-night working and the tone of their own notes. In the demo data it flags Ramesh Patel at 85/100 — critical, before anybody complained.
It reads the audit chain, not the task list — because somebody who edits a record cannot edit the evidence that they edited it. In the demo it catches an officer who approved his own work and closed a field inspection in 4 minutes.
92% precision — 11 of 12 flagged cases were genuine on review. Not 100%, because one was a real false alarm: a man working late who turned out to be overloaded, not dishonest. We label it as a miss instead of quietly deleting it.
Ask "mere pending kaam kitne hain?" and it answers from your real tasks. It cannot make up a number, because it never writes numbers — it picks the answer shape and fills in figures straight from the database.
- Works in Hindi. One click swaps the whole interface. Task text stays as written — translating a citizen's file note would misrepresent the record.
- New staff can register themselves. Sign up → verify by email code → an administrator approves → you're in. Self-registration can only ever create a normal employee account, never a manager.
- Nobody approves their own work. The system blocks it, whatever your rank.
- Drag-and-drop task board for managers, with illegal moves refused.
- Reports you can download as spreadsheets.
- Works on a phone and installs like an app.
- Built for accessibility — full keyboard use, screen-reader labels, and all animation switches off if your device asks for reduced motion.
Every number below was measured on the demo data and is shown live in the app, not typed into a slide.
| Claim | Measured | Where to check |
|---|---|---|
| Work gets done 30–40% faster | 32.5% | Org Overview, top-right card |
| Mood detection is 85–90% accurate | 87.5% | npm run eval:ml |
| Fraud detection ~92% precise | 11 of 12 | Fraud & Risk Center |
| Records are tamper-evident | 962 blocks, 15 ms | Blockchain Audit → Verify |
| 30+ documented APIs | 66 | http://localhost:4000/docs |
What is simulated, stated plainly:
- The Parichay government login screen is a practice version. The real one needs official NIC approval. Every screen says Sandbox.
- Blockchain anchoring is designed, not built. No cryptocurrency, no wallet, no transaction.
- The AI runs offline by default so a demo cannot fail on bad wifi.
| Layer | Technology |
|---|---|
| Website | Next.js 14, TypeScript, Tailwind CSS, Framer Motion |
| Server | Node.js, Express, Prisma, PostgreSQL 16 |
| AI | Python, FastAPI, scikit-learn, DistilBERT |
| Security | SHA-256 hash chain, JWT sign-in, bcrypt passwords |
| File | What's in it |
|---|---|
| DEMO.md | A rehearsed 7-minute script for showing this to judges |
| TECHNICAL.md | Full technical reference — API, models, evaluation method |
| DECISIONS.md | Every engineering decision and why, including the mistakes |
Smart India Hackathon prototype · Aligned with the Digital India initiative
The circular emblem is a generic departmental monogram. The State Emblem of India is deliberately not used — its use is restricted by law.