EcoLafaek - Autonomous AI Agent for Environmental Monitoring

EcoLafaek demonstrates the power of Amazon Bedrock AgentCore to create truly autonomous AI agents that solve real-world environmental challenges in Timor-Leste. Our system showcases multi-round tool calling, code execution, browser automation, and intelligent decision-making through advanced reasoning LLMs.

✅ LLM from AWS Bedrock: Amazon Nova-Pro v1.0 (amazon.nova-pro-v1:0)

✅ Amazon Bedrock AgentCore:

• ✅ Code Interpreter primitive for autonomous chart generation
• ✅ Browser Tool primitive for web scraping
• ✅ Application runs on agentcore_app.run() framework

✅ Autonomous AI Agent:

• ✅ Uses reasoning LLM (Nova-Pro) for decision-making
• ✅ Demonstrates autonomous capabilities with multi-round tool calling (up to 5 rounds)
• ✅ Integrates external tools: SQL databases, code execution, web scraping, S3 storage

✅ Production Deployment: Live on AWS Lightsail

🤖 Agent Innovation Highlights:

• Multi-Round Tool Orchestration: Agent autonomously chains SQL → Chart Generation → Map Creation → Web Scraping
• Code Interpreter Integration: Generates matplotlib/pandas visualizations on-demand via AgentCore
• Browser Automation: Scrapes web content using Playwright via AgentCore Browser Tool
• Intelligent Decision-Making: Nova-Pro reasoning determines which tools to call and in what sequence
• Real-World Impact: Solving waste management crisis affecting 300+ tons daily in Timor-Leste

---

🚀 Judge Quick Start

🎯 Live Demo Access:

Component	URL	Credentials
📱 Mobile App	Download APK	Username: usertest Password: 1234abcd
🌐 Public Dashboard	www.ecolafaek.com	No login required
🤖 AI Agent Chat	www.ecolafaek.com/agentcore-chat	Try: "Show waste trends chart"
⚡ Backend API	www.ecolafaek.xyz/health	Health check endpoint

📚 Complete Documentation:

https://docs.ecolafaek.com

Component	Documentation	Description
📐 Architecture	Diagram/README.md	Complete system architecture
⚡ Backend API	mobile_backend/README.md	AgentCore implementation details
🌐 Dashboard	ecolafaek_public_dahboard/README.md	Frontend integration
📱 Mobile App	ecolafaek/README.md	Flutter mobile client
🗄️ Database	database/README.md	Schema and vector storage

---

🌟 About EcoLafaek

EcoLafaek (named after the crocodile "Lafaek" in Timorese culture) is an AI-powered environmental monitoring system that empowers citizens of Timor-Leste to combat waste management challenges through intelligent reporting and autonomous data analysis.

🎯 The Problem

Timor-Leste's capital Dili faces a severe waste crisis:

• 300+ tons of waste generated daily
• 100+ tons go uncollected each day
• Blocked drainage systems cause flooding during rainy season
• Limited infrastructure and resources for waste management

Source: JICA Survey on Solid Waste Management

💡 Our Solution

An autonomous AI agent system that:

1. Analyzes waste images using Amazon Bedrock Nova-Pro multimodal LLM
2. Classifies waste types and severity automatically
3. Generates real-time analytics and visualizations via AgentCore Code Interpreter
4. Provides intelligent insights through natural language chat interface
5. Empowers communities with data-driven decision making

---

🏗️ System Architecture

Core Components:

Detailed Architecture: See Diagram/README.md

---

🤖 Autonomous AI Agent Workflow

Multi-Round Tool Calling Example

User Query: "Show me waste trends and create a map of hotspots"

┌─────────────────────────────────────────────────────────────┐
│ Round 1: Nova-Pro Reasoning                                 │
│ → "I need to get waste data first"                          │
│ → Calls: execute_sql_query                                  │
│   SELECT DATE(created_date), waste_type, COUNT(*)           │
│   FROM reports GROUP BY DATE(created_date), waste_type      │
└─────────────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────────────┐
│ Round 2: Nova-Pro Reasoning                                 │
│ → "Got the data, now create a trend chart"                  │
│ → Calls: generate_visualization                             │
│   AgentCore Code Interpreter executes Python:               │
│   - import matplotlib.pyplot as plt                         │
│   - Generate line chart                                     │
│   - Return base64 PNG → Upload to S3                        │
└─────────────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────────────┐
│ Round 3: Nova-Pro Reasoning                                 │
│ → "Now get hotspot locations for the map"                   │
│ → Calls: execute_sql_query                                  │
│   SELECT name, center_latitude, center_longitude,           │
│   total_reports FROM hotspots WHERE status='active'         │
└─────────────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────────────┐
│ Round 4: Nova-Pro Reasoning                                 │
│ → "Create an interactive map with hotspot markers"          │
│ → Calls: create_map_visualization                           │
│   Generates Folium HTML map → Upload to S3                  │
└─────────────────────────────────────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────────────┐
│ Round 5: Final Response                                     │
│ → Returns markdown with:                                    │
│   - Chart image: ![Trend](s3_url)                          │
│   - Interactive map link                                    │
│   - Data analysis summary                                   │
└─────────────────────────────────────────────────────────────┘

Example Log

Agent Tools Available

Tool Name	AgentCore Component	Purpose	Example
execute_sql_query	Direct Execution	Database queries	"How many reports last week?"
generate_visualization	Code Interpreter	Chart generation	"Show waste distribution chart"
create_map_visualization	Code Interpreter	Map creation	"Map hotspots"
scrape_webpage_with_browser	Browser Tool	Web scraping	"What is EcoLafaek?"
get_ecolafaek_info	Browser Tool	Project info	"Tell me about features"

Example Chat

---

📱 Components Overview

1. Mobile Application (Flutter)

• Cross-platform iOS/Android app
• Camera integration for waste photo capture
• GPS location tracking
• Real-time AI analysis feedback
• Personal impact dashboard

→ Full Mobile App Documentation

2. Public Dashboard (Next.js + Vercel)

• AI Agent chat interface with multi-round tool calling
• Semantic vector search powered by Amazon Titan Embed
• Interactive geospatial maps
• Real-time analytics and reporting
• Community leaderboards

→ Full Dashboard Documentation

3. Backend API (FastAPI + AgentCore)

• Runs on agentcore_app.run() framework
• Amazon Bedrock Nova-Pro integration for reasoning
• AgentCore Code Interpreter for chart generation
• AgentCore Browser Tool for web scraping
• Multi-round conversational AI with tool orchestration
• Image analysis with Amazon Bedrock invoke_model()
• Deployed on AWS Lightsail

→ Full Backend Documentation

4. Database (Distributed SQL + Vectors)

• User authentication and management
• Waste report storage with GPS coordinates
• AI analysis results with 1024-dim vector embeddings
• Hotspot detection and clustering
• Multi-application access

→ Full Database Documentation

5. Admin Panel (Next.js - Local Only)

• User management and moderation
• Report oversight and analytics
• System configuration
• AI performance monitoring
• Security: Not deployed publicly, local access only

→ Full Admin Panel Documentation

---

🚀 Getting Started

For Judges - Quick Testing

1. Try the Live Dashboard:

   Visit: https://www.ecolafaek.com
   Click: "Agent Chat" → Ask: "Show waste type distribution chart"

2. Test Mobile App:

   Download: https://www.ecolafaek.com/download
   Login: usertest / 1234abcd
   Try: Submit a report with photo

3. Explore Vector Search:

   Visit: https://www.ecolafaek.com/vector-search
   Enter: "plastic waste pollution"
   See: Semantic similarity results

For Developers - Local Setup

See component-specific README files for detailed setup instructions:

• Backend Setup
• Dashboard Setup
• Mobile App Setup

---

🎯 Technical Highlights

Amazon Bedrock Integration

Nova-Pro LLM (amazon.nova-pro-v1:0):

• Multi-modal image + text analysis
• Autonomous reasoning and decision-making
• Tool orchestration and planning
• Multi-round conversational capabilities

Titan Embed (amazon.titan-embed-image-v1):

• 1024-dimensional vector embeddings
• Semantic similarity search
• Image and text embedding generation

AgentCore Primitives

Code Interpreter:

with code_session(region='us-east-1') as client:
    result = client.invoke('executeCode', {
        'language': 'python',
        'code': chart_generation_code
    })

Browser Tool:

with browser_session(region='us-east-1') as client:
    ws_url, headers = client.generate_ws_headers()
    browser = playwright.chromium.connect_over_cdp(ws_url, headers=headers)

---

📊 Impact & Scale

• ✅ Production Deployment: Live system with real users
• 📱 3+ Active Users: Mobile app downloads and engagement
• 🗺️ 50+ Reports: Waste reports submitted and analyzed
• 🤖 25+ AI Interactions: Agent tool executions

---

📹 Demo Video

🛠️ Kiro - AI-powered Integrated Development Environment (IDE)

During the development of EcoLafaek, we leveraged Kiro - AWS's AI agent platform - to accelerate our development workflow and enhance code quality.

How Kiro Enhanced Our Development:

🔍 Code Analysis & Review

• Analyzed complex AgentCore integration code for best practices
• Identified potential bugs and security vulnerabilities
• Suggested optimizations for multi-round tool calling logic

📊 Architecture Diagram Generation

• Generated system architecture diagrams from codebase
• Created visual representations of agent workflow
• Helped document complex AI agent interactions

🤖 AI Agent Code Assistance

• Guided implementation of AgentCore Code Interpreter integration
• Provided examples for Browser Tool usage with Playwright
• Assisted with Amazon Bedrock API integration patterns

📚 Documentation Generation

• Helped structure comprehensive README files
• Generated API documentation from code comments
• Created deployment guides and setup instructions

⚡ Rapid Prototyping

• Accelerated development of visualization tools
• Streamlined database schema design
• Quick testing of different AI prompt strategies

Kiro's autonomous capabilities allowed us to focus on solving Timor-Leste's waste management challenges while maintaining high code quality and comprehensive documentation.

---

🌿 Built with ❤️ for Timor-Leste 🌿

AWS AI Agent Global Hackathon

Powered by Amazon Bedrock AgentCore, Nova-Pro, and Titan Embed

"Lafaek" - The Sacred Crocodile Guardian of Timor-Leste