Tatkal Copilot
Stop Guessing. Start Winning Tatkal.
An AI travel agent that prepares you before the Tatkal window opens. It never
books, logs in, or automates anything — it makes you dramatically better
prepared, then hands you off to the official IRCTC site.
Built for the Build What Moves India hackathon (OpenAI + Varun Mayya).
The one user
Everything is designed for Manoj Sharma, 54, Mumbai — books Tatkal 3–5×/year,
not tech-savvy, usually relies on his son or a travel agent. His goal:
"I need to reach Delhi before 8AM tomorrow."
The one journey
A single continuous story across 8 screens:
- Hero — the promise
- Travel Goal — plain words, speech input ("reach Delhi before 8AM with my wife")
- AI Thinking — visible reasoning, not an instant answer
- Strategy — one confident recommendation (August Kranti, 63%), why not the
obvious train, the Borivali boarding edge, and a Split-via-Kota backup
- Passenger Vault — saved passengers, one tap
- Mission Control — countdown, live readiness checklist, streaming AI coach
- Booking Assistant — security handoff to the official IRCTC site (new tab)
- Demo Completion — measurable before/after (28% → 63%, 0 → 2 backups)
How the AI is used (meaningfully, and safely)
The AI does the two things it is genuinely good at:
- Understands Manoj's messy sentence → structured intent.
- Explains the strategy like a seasoned travel agent.
Every number and the strategy itself stay grounded in mock data via a
deterministic local planner (lib/planner.ts), so GPT can never invent a fake
confirmation figure.
Dual-path resilience: with OPENAI_API_KEY set, the /api/plan route uses
GPT (gpt-4o-mini). Without a key — or if the call fails — it falls back to the
local planner and produces the same shape of expert output. The demo never
breaks in front of judges.
Run it
npm install
npm run dev
Open http://localhost:3000.
To enable the live GPT path:
cp .env.example .env.local # add your OPENAI_API_KEY
Stack
Next.js 16 · React 19 · TypeScript · Tailwind v4 · Framer Motion · Lucide.
Local JSON data, no database, no auth, no backend beyond one API route.
Architecture
app/ layout, page orchestrator, /api/plan (GPT + fallback)
components/ ui primitives (button, card, probability ring), brand chrome
features/ one folder per screen (hero, goal, thinking, strategy, …)
lib/ planner (intelligence), ai (dual-path), i18n (EN/HI), journey state
data/ trains, stations, split routes, passengers, mission script (JSON)
hooks/ countdown, speech
types/ domain model
Principles
- Never pretends to connect to IRCTC. Never automates booking. Never scrapes.
- Every screen answers: "How does this reduce Tatkal anxiety?"
- Accessibility: WCAG-minded contrast, large targets, EN/हिन्दी toggle, reduced-motion.
