IslandFlow (Weather-intelligent marine dispatch)

Persistent guest memory meets weather-intelligent marine dispatch, powered by Qwen 3.7 Plus and backed by Alibaba Cloud OSS.

Comment

Inspiration

The idea for IslandFlow was born in the tropical archipelagos of Bocas del Toro, Panama—home to some of the world’s most exclusive overwater luxury resorts (like Nayara, Sweet Bocas, and La Coralina). In these remote paradise destinations, life revolves entirely around the water. Guests travel by boat, dine by boat, and book outdoor excursions like snorkeling, surfing, and island-hopping.

But paradise has a major operational bottleneck: unpredictable tropical weather.

A sudden storm downpour or a sudden wave-height spike can instantly ruin an outdoor excursion. In boutique hospitality, handling these disruptions is a manual nightmare. Disappointed guests clog the front desk with frantic calls, operators spend hours calling captains and manually rescheduling itineraries, and boat captains are left stranded in hazardous waters. We wanted to build a smart, empathetic, and weather-intelligent bridge; an agentic concierge that seamlessly orchestrates real-time weather forecasts, resort databases, and guest preferences to automate scheduling and marine dispatches in a single, gorgeous dashboard.

What it does

IslandFlow is a weather-intelligent, multi-tenant guest concierge and marine dispatch platform designed for luxury boutique resorts. It splits the experience into two unified, real-time views:

The Guest Concierge Experience: Guests get a premium, mobile-responsive dashboard that breathes with the island. Guests can chat with an agentic local concierge to ask questions, explore the island, or book activities. The Proactive Rescheduling Loop: When a real-world storm or high-wave alert triggers, IslandFlow doesn't wait for things to go wrong. It scans the database for active outdoor bookings on the affected date, identifies conflicts, and queries the AI to match that specific guest's preferences with alternative indoor activities (like cocoa farm workshops or couples' massages). It then renders an interactive "Rescheduling Swap Card" directly inside the guest’s active chat, letting them approve the new booking in 1-tap and writing it back to MongoDB Atlas in real time. The Operator & Dispatch Console: Front-desk dispatchers get a powerful control center. They can monitor real-time captain dispatches, oversee guest profiles, simulate real-world weather states, sync guest data via a Property Management System (PMS) API, and dynamically onboard completely new hotel brands with custom color palettes.

How we built it

We built IslandFlow to demonstrate how modern, cloud-native orchestration can solve complex real-world logistics:

The Brain & AI Orchestration: Built entirely on Qwen Cloud utilizing the cutting-edge Qwen3.7-Plus model. We engineered a custom Qwen Orchestrator framework to parse complex natural language guest intents, coordinate state transitions, query database models, and write finalized itineraries.

Hosting & Infrastructure: Deployed on an Alibaba Cloud ECS (Elastic Compute Service) instance in a scalable Docker container, ensuring high-availability routing for our backend and frontend layers.

Backend: Created a highly performant FastAPI (Python) server utilizing asynchronous background workers to handle periodic real-world weather updates (using the OpenWeather API) and trigger proactive rescheduling campaigns.

Database: Powered by MongoDB Atlas to securely manage multi-tenant hotel profiles, guest bookings, and captain dispatch queues.

Frontend: Developed a stunning, high-fidelity React client. We bypassed bloated modern CSS frameworks in favor of handcrafted, lightweight vanilla CSS to achieve silky-smooth glassmorphism, responsive micro-animations, and a premium sliding feedback module.

Direct Alibaba Cloud API Usage (OSS): Integrated the official Python oss2 SDK to compile finalized itineraries as beautiful, official travel receipts and upload them directly to an Alibaba Cloud OSS bucket, returning signed, secure URLs.

Proof of Alibaba Cloud API Usage & Deployment

To satisfy the official requirement of demonstrating direct usage of Alibaba Cloud services and live deployment, we have integrated the Alibaba Cloud Object Storage Service (OSS) API and hosted the entire system on Alibaba Cloud compute:

  • Live Deployment Instance: http://47.82.220.6/ (Running live on an Alibaba Cloud ECS Server)
  • Direct API/SDK Usage Proof: backend/ali_oss.py (Demonstrates direct programmatic interaction using the official Alibaba Cloud oss2 SDK library)

Challenges we ran into

LLM Intent Precision in Multi-Agent Workflows: Getting standard AI models to reliably read database states, check excursion capacities, verify guest slots, and generate structured JSON rescheduling payloads without hallucinating was incredibly difficult. We resolved this by building a strict deterministic controller inside our Qwen Orchestrator on Qwen Cloud, wrapping all database writes inside strict Pydantic schemas and structured tool calls. Responsive Styling without Tailwind CSS: Achieving a premium, elite-agency aesthetic using only standard vanilla CSS was a massive layout challenge. We hand-coded custom CSS grid parameters, theme transitions, and sliding drawer coordinates to keep the compiled bundle exceptionally light and lightning-fast. Non-Intrusive Real-Time Synchronization: During status polling, if an itinerary updated in the background, a full-screen modal would initially pop up and interrupt the guest mid-sentence. We spent significant time isolating states, ensuring that background updates seamlessly refresh inline on the guest's dashboard, keeping full-screen popups reserved strictly for explicit, user-initiated actions. Accomplishments that we're proud of 100% Native Qwen Cloud Architecture: We successfully built and ran a complex, multi-agent hospitality platform entirely on the Qwen Cloud ecosystem, proving that it is fully capable of driving world-class agentic products. Stunning Visual Aesthetics: We designed a UI that looks and feels like an expensive, custom-built boutique SaaS product that a five-star resort would proudly put in front of high-paying guests. Autonomous, End-to-End Rescheduling: We succeeded in making an AI that can autonomously detect a weather threat, match a guest to an indoor alternative, render a beautiful card, write it to MongoDB Atlas, and generate a signed receipt on Alibaba Cloud OSS in under 3 seconds.

What we learned

Agentic AI is Proactive, Not Passive: Large language models are no longer just conversational text engines. When connected to robust, real-time databases and third-party APIs (like weather or PMS systems), they become highly effective, autonomous coordinators of complex physical logistics. UX Empathy is Everything: Building a system with two highly distinct users—luxury resort guests seeking zero stress, and busy front-desk dispatchers managing marine assets—taught us how to design interfaces that hide immense backend complexity behind beautiful, simple, 1-tap layouts.

What's new

While the project was started after May26th, it has been submitted to other hackathons within the period of this hackathon. The key differences with the previous submissions are:

During this hackathon cycle, IslandFlow was thoroughly overhauled from its previous template/concierge iterations and enriched with several major, production-grade features to transition it from a standard booking manager to a weather-intelligent, relationship-driven MemoryAgent natively integrated with Alibaba Cloud.

The core differences and additions made during this period include:

  1. *Unified Core Cognitive Qwen Brain * (Migration from AWS/Gemini to Qwen Studio) Previously: Used a generic LLM setup (Gemini/OpenAI) on basic prompts to manage bookings. Now: Replaced the entire core reasoning stack with Qwen 3.7 Plus via the Alibaba Cloud DashScope API. The system prompts, token budgets, and tool definitions are customized specifically around Qwen's schema rules, ensuring rapid, zero-shot conversational reasoning, bilingual translation, and accurate function-calling.
  2. *Persistent Cross-Session Memory * (Track 1: MemoryAgent) Previously: Conversational context was lost on page refresh, and user preferences were stored inside ephemeral, single-session states. Now: Created a robust persistent memory module using MongoDB. When guests chat with the concierge (e.g., "I have a severe shellfish allergy and prefer private excursions"), Qwen autonomously invokes the custom save_conversational_memory tool. Upon page reloads or profile switches, these records are pre-fetched and injected directly into the Qwen agent's background prompt block, allowing it to remember critical health and activity preferences across separate sessions without repeating expensive database calls.
  3. Native Alibaba Cloud Object Storage Service (OSS) Integration Previously: Did not interact with any Alibaba Cloud services or SDKs (the project was hosted and formatted with basic local file outputs). Now: Added full-stack integration with Alibaba Cloud OSS using the official Python oss2 SDK. When a guest's itinerary is updated, the backend compiles a formatted travel document and streams it dynamically as a byte file to a Singapore-based Alibaba OSS bucket. It then returns a secure, signed public download URL valid for 24 hours, furnishing verifiable in-repo proof of Alibaba Cloud API usage.
  4. Dynamic Weather Synchronization (Open-Meteo Integration) Previously: Swapped weather using simple static UI toggles that didn't sync to the database or model backend. Now: Implemented a live Open-Meteo weather syndication service. The backend dynamically fetches, calculates, and synchronizes average daily marine weather variables (temperatures, precipitation, and wave heights) directly to MongoDB.
  5. Dynamic Captain Onboarding & Fleet Expansion Flow Previously: Sea captains and boat assignments were hardcoded in a static Mock JSON database. Now: Added a complete onboarding flow. Resort operators can now register new local water-taxi or excursion captains directly from the control panel. The frontend submits a POST request to a new FastAPI /api/operator/add-captain endpoint, writing to a persistent MongoDB captains collection and refreshing the live operator assignable state instantly.

What's next for IslandFlow (Weather-intelligent marine dispatch)

Deep Production PMS Integrations: Hooking up our PMS-Sync endpoint to live hospitality platforms like Mews, Opera, and Cloudbeds so boutique hotels can install IslandFlow with a single click. Telemetry & IoT Captain Dispatch: Connecting real-time GPS Beacons to the dispatch boats so that captain notifications on the dispatcher's dashboard are updated by live maritime telemetry. Qwen Low-Latency Voice Concierge: Integrating Qwen Cloud’s voice capabilities so guests can talk to the concierge naturally from their smartphones while relaxing on the beach.

Built With

  • agentic-concierge
  • alibaba-cloud
  • alibaba-cloud-ecs
  • antigravity
  • automated-rescheduling
  • css3
  • docker
  • dynamic-theme-switcher
  • fastapi
  • glassmorphism
  • gsap
  • html5
  • javascript
  • mongodb-atlas
  • multi-agent-systems
  • pms-integration
  • python
  • qwen-cloud
  • qwen-orchestrator
  • qwen-plus
  • react
  • svg-vectors
  • vite
  • weather-intelligent
  • websockets
Share this project:

Updates