Project Story: Taawun Maroc (تعاون المغرب)

About the Project

Taawun Maroc is a multi-agent AI platform designed to coordinate humanitarian aid in post-disaster Morocco, especially after earthquakes or floods. Citizens report urgent needs (food, medicine, shelter, volunteers) via text, photos, and GPS in Darija, Arabic, French, or Tamazight. Gemini 3-powered agents process everything: deduplication, prioritization, volunteer matching, and synthesis of bilingual reports for authorities. An interactive heatmap visualizes needs in real-time, embodying Morocco's spirit of taâwun (mutual aid) through modern AI.

What Inspired Me

Living in Casablanca, I witnessed the 2023 Al Haouz earthquake's chaos: duplicated aid requests, delayed medical help, and poor coordination between communities and authorities. I wanted to empower citizens to become active in relief while bridging the gap to institutions. The recent Gemini 3 release—with its advanced multimodal, long-context reasoning, and agentic capabilities—inspired me to build a truly adaptive, culturally-aware system.

How I Built It

I used Antigravity (Google's agentic IDE) to bootstrap the MVP in hours.

• Frontend: React + Vite + Tailwind for a responsive UI with Leaflet heatmap.
  
• Backend: Node.js + Express + SQLite for storing normalized needs.
  
• Core AI: Gemini 3 Flash API (multimodal for photo analysis, long-context for handling 20+ concurrent needs, reasoning for prioritization and matching).
  Agents are chained via prompt engineering: ingestion → deduplication → urgency scoring (with Morocco-specific factors like seismic vulnerability) → skill/location matching → report generation. Mock data simulates realistic scenarios in Marrakech, Taroudant, and Al Haouz.

Gemini Integration (Core Focus)

Gemini 3 is central:

• Multimodal capabilities analyze damage photos + text to extract need type, location, urgency.
  
• Long-context reasoning processes dozens of posts simultaneously for global prioritization and deduplication via semantic understanding.
  
• Advanced agentic workflows enable collaborative multi-agents (handoffs via CoT prompting).
  
• Built-in multilingual support handles Darija nuances without extra translation APIs.
  This makes the system fast, accurate, and culturally relevant—key for real impact.

What I Learned

Gemini 3's reasoning depth surprised me: it captures subtle urgency cues (e.g., "enfants blessés" vs "faim") better than rule-based systems. Prompt engineering for Darija required careful examples, but yielded strong results. I learned vibe-coding speed: from idea to demo in <24h thanks to Antigravity.

Challenges Faced

• Tight hackathon deadline (sprinted last hours!).
  
• Simulating realistic multilingual/Darija inputs without large datasets.
  
• Balancing MVP simplicity vs. wow-factor visuals (heatmap + agent flow animation).
  
• API rate limits/quota management during intense testing.
  

Despite the rush, Taawun Maroc proves AI can amplify human solidarity in crises. Proud to build something meaningful for Morocco.

Built With

• advanced-agentic-workflows-via-chain-of-thought-prompting-for-multi-agent-collaboration
• agent-orchestration
• al-haouz
• and-data-storage.-**database**:-sqlite-?-lightweight-in-memory-storage-for-mvp-(normalized-needs
• and-self-verification.-**other**:-google-generative-ai-sdk-(@google/generative-ai)-for-seamless-gemini-3-integration.-gemini-3-is-central-and-powers-the-entire-intelligent-coordination-loop?without-it
• arabic
• built-in-multilingual-support-(darija
• built-with**-**gemini-3-api**-(gemini-3-flash-preview)-?-core-intelligence:-multimodal-analysis-(text-+-images-for-damage-photos)
• code-generation
• french
• iteration
• long-context-reasoning-for-deduplication/prioritization-of-multiple-needs
• marrakech
• mock-data).-**mapping**:-leaflet.js-?-interactive-map-+-heatmap-layer-to-display-prioritized-needs-(e.g.
• tamazight-nuances).-**frontend**:-react-+-vite-+-tailwind-css-?-responsive-ui-with-interactive-leaflet/openstreetmap-heatmap-for-visualizing-needs-across-morocco-regions.-**backend**:-node.js-+-express-?-api-server-handling-posts
• taroudant-post-earthquake-scenarios).-**development-tool**:-google-antigravity-(agentic-ide)-?-used-to-bootstrap-the-full-mvp-autonomously-with-vibe-coding:-architecture-planning
• the-multi-agent-system-wouldn't-handle-real-time
• volunteer-profiles