No sign-up needed — there's a guest mode. Open it on a phone if you can, it's built handset-first.

TravelBuddy

A booking ends at checkout. The trip doesn't.

An AI trip concierge for Goa — it plans the days, books the table, and then stays with you on Telegram for the whole trip.

_{Built for Geeks2Code 2026 · Wayzyy special track · Team AlphaForge}

Every travel app in India stops working the moment you've paid. You land in Goa with a hotel confirmation and a group chat, and everything that actually makes the trip good — where to eat tonight, whether the waterfall is worth the two-hour drive, what to do now that it's raining — is back to being your problem.

TravelBuddy covers the part after checkout. It learns what you like by watching you swipe, builds a costed day-by-day plan you can edit, books you into a partner network, and then follows you into Telegram so the plan is a message away for the rest of the trip.

Swipe to be understood 21 cards — 7 vibes, 7 activities, 7 food — over real Goa photography. No forms, no "rate your interest from 1 to 5". Pure-veg filters the deck. Swipe left on everything and it puts two options head-to-head instead of giving up. You can also just point your camera at a photo you like; a vision model reads the scene and pre-seeds your preferences.
	A plan you can argue with Day-by-day, timed, costed against the budget split you set. Every stop can be edited, moved or replaced with a real Goa alternative. Live demand pricing moves costs through the day and tells you when something is worth rebooking. Export the whole thing as a PDF to send to whoever isn't on the app.
It books, and it earns Confirm a plan and the guest earns Travel Cash — credit that only spends at partner venues, which is what pulls them into the network instead of to whoever happens to be nearest. Every food, activity and transport stop also surfaces bookable partners with prices, ratings and one-tap call or WhatsApp. That's the business model, not a feature: TravelBuddy takes 8–25% commission on what the itinerary funnels into, and a "Partner economics" toggle on each stop shows the unit economics per booking.
	Then it leaves the app Link your plan to Telegram in one tap and the concierge follows you there. It nudges you 45 minutes before each stop, flags price moves, answers questions about your own itinerary, and knows the full partner network — so "where should we eat tonight?" gets an answer you can actually book. Share the link and your travel companions get the same bot, same plan.

Every surface is themed through design tokens, contrast-checked by measurement rather than by eye, and follows the system setting unless you override it in your profile.

Message @wayzyy_goa_bot, or link it from the app to bind it to your trip.

Ask it "is Dudhsagar worth it on day 3?" or "cheap dinner near Anjuna" and it answers from your itinerary and the partner network — not from generic internet knowledge about Goa.

flowchart TB
    subgraph client["Browser / PWA / Android TWA"]
        UI["Next.js 15 App Router<br/>React 19 · Tailwind v4"]
        LS[("localStorage<br/>trip · planner · history")]
        UI <--> LS
    end

    subgraph edge["Vercel serverless"]
        API["/api routes"]
        ASSIST["lib/assistant.ts<br/>one prompt, two surfaces"]
        TG["/api/telegram<br/>webhook"]
        CRON["/api/cron/notify<br/>45-min lookahead"]
        API --> ASSIST
        TG --> ASSIST
    end

    subgraph store["State the bot needs"]
        KV[("Upstash Redis<br/>snapshots · tokens · chat links")]
    end

    subgraph ext["External"]
        GROQ["Groq<br/>gpt-oss-120b"]
        HF["Hugging Face<br/>gemma-3-4b-it vision"]
        OSM["OpenStreetMap<br/>Nominatim · tiles"]
        TOMTOM["TomTom routing"]
        TGAPI["Telegram Bot API"]
    end

    UI -->|"plan, chat, photo"| API
    UI -->|"link trip"| API
    API --> KV
    TG --> KV
    CRON --> KV
    ASSIST --> GROQ
    API --> HF
    UI --> OSM
    API --> TOMTOM
    TG <--> TGAPI
    CRON --> TGAPI
    GHA["GitHub Actions<br/>every 10 min"] --> CRON

Three things worth knowing:

1. Trip state lives in the browser. There is no user database. The app persists to localStorage per signed-in email, so a trip belongs to a device until you link it out.
2. One assistant, two surfaces. src/lib/assistant.ts holds the Groq call and the prompt. The in-app support chat and the Telegram bot both call ask(), so the itinerary changes once and every surface sees it.
3. Transport is an interface. Notifier in src/lib/telegram.ts abstracts the channel — WhatsApp drops in as a second adapter without touching the assistant.

.
├── .github/workflows/
│   └── trip-notifications.yml   # 10-min sweep → /api/cron/notify
└── travelbuddy/
    ├── travel-buddy/            # the Next.js app — everything live
    │   ├── src/app/             # routes + API handlers
    │   ├── src/components/      # travel/, itinerary/, onboarding/, auth/
    │   ├── src/lib/             # assistant, telegram, store, pricing, schedule
    │   ├── src/data/            # vendors, sponsors, Goa knowledge base
    │   ├── public/goa/          # 55 photographs
    │   └── SKILL.md             # working notes, read this before contributing
    └── server/                  # Node/Express + Gemini agent — NOT deployed

git clone https://github.com/raunak23427/travel-concierge.git
cd travel-concierge/travelbuddy/travel-buddy
npm install
cp .env.example .env.local   # then fill it in, see below
npm run dev

Open http://localhost:3000. The app works with no keys at all — it falls back to a local Goa knowledge base — but the assistant, photo analysis and Telegram bot need theirs.

npm run typecheck   # tsc --noEmit
npm run build       # production build
npm run test:travel # Playwright

This is a hackathon prototype, and the README should say which parts are real:

• ✅ Real: the full plan → swipe → itinerary → confirm flow, the Telegram bot end to end, the Groq assistant, photo analysis, dark mode, PDF export, Google sign-in, durable bot state on Upstash.
• ⚠️ Sample data: the 18 partner vendors are real Goa venues with realistic prices, but the phone numbers are placeholders and no partner agreement exists. Nothing here is a live booking network.
• ❌ Not deployed: travelbuddy/server/ — the Node/Express agent. The web app falls back to its local knowledge base and does not call it.

AlphaForge — Geeks2Code 2026, Wayzyy special track.