Vikram30069/RescueNet-AI

## 📋 Table of Contents - [🌋 The Problem](#-the-problem) - [⚡ The Solution](#-the-solution) - [🏗️ System Architecture](#️-system-architecture) - [🤖 The 10-Agent AI Pipeline](#-the-10-agent-ai-pipeline) - [☁️ AWS Infrastructure](#️-aws-infrastructure) - [📱 Twilio Automation Layer](#-twilio-automation-layer) - [🗺️ Real Research Datasets — Telangana](#️-real-research-datasets--telangana) - [🛠️ Tech Stack](#️-tech-stack) - [🚀 Quick Start](#-quick-start) - [📂 Repository Structure](#-repository-structure) - [🔌 API Reference](#-api-reference) - [⚙️ Configuration](#️-configuration) - [📚 Key Documents](#-key-documents) - [🗺️ Roadmap](#️-roadmap) ## 🌋 The Problem When disasters strike — **floods, earthquakes, fires, industrial accidents** — emergency responders face an overwhelming information crisis: | Pain Point | Impact | |---|---| | 🔴 **Siloed Communication** | Fire, medical, and police systems operate on separate channels → delayed coordination | | 🔴 **Cognitive Overload** | Human commanders process thousands of fragmented reports → critical decision delays | | 🔴 **Capacity Blindness** | Teams dispatched without real hospital bed data → transport bottlenecks | | 🔴 **No Survivor Scoring** | No quantitative model to rank who needs help first → rescue by guesswork | | 🔴 **Manual Alert Drafting** | Operators hand-write SMS and call scripts under extreme stress → errors, delays | ## ⚡ The Solution RescueNet AI is a **fully automated multi-agent orchestration system** that transforms a raw incident report into a complete, dispatched rescue plan — entirely through AI collaboration. 📡 INCIDENT REPORTED ↓ (seconds) 🧠 10 AI AGENTS COLLABORATE IN SEQUENCE ↓ 📊 SURVIVOR RISK SCORES GENERATED ↓ 🚒 HOSPITALS & RESOURCES MATCHED FROM REAL DATASETS ↓ 📋 STRUCTURED RESCUE PLAN JSON PRODUCED ↓ 📱 AUTOMATED ALERTS DISPATCHED (SMS · WhatsApp · Voice · Dashboard) **End-to-End Time**: `< 1 second (mock)` | `10–45 seconds (live LLM)` ## 🏗️ System Architecture ┌──────────────────────────────────────────────────────────────────────────────────┐ │ RESCUENET AI — FULL SYSTEM ARCHITECTURE │ └──────────────────────────────────────────────────────────────────────────────────┘ 👤 FIELD TEAMS / COMMANDERS / DISPATCHERS │ Browser / Mobile / API ▼ ┌────────────────────────────────────────────────────────────────────────────┐ │ AWS AMPLIFY (CDN) │ │ Next.js 14 Frontend — App Router / TypeScript │ │ Dashboard · Incident Form · Live Map · Rescue Plan Viewer │ └──────────────────────────────────────┬─────────────────────────────────────┘ │ REST API (HTTPS) ▼ ┌────────────────────────────────────────────────────────────────────────────┐ │ AMAZON EC2 + DOCKER + ECR │ │ FastAPI Backend — Python 3.11 — Pydantic v2 │ │ │ │ ┌───────────────┐ ┌──────────────────┐ ┌────────────────────────┐ │ │ │ REST Routers │ │ Service Layer │ │ DB Repository Layer │ │ │ │ /incidents │ │ (Orchestration │ │ PostgreSQL / Neon │ │ │ │ /agents │ │ & Business │ │ · incidents │ │ │ │ /hospitals │ │ Logic) │ │ · rescue_requests │ │ │ │ /resources │ │ │ │ · agent_decisions │ │ │ │ /rescue-plan │ │ │ │ · alert_logs │ │ │ └───────────────┘ └──────────────────┘ └────────────────────────┘ │ └──────────────────────────────────────┬─────────────────────────────────────┘ │ Python function call ▼ ┌────────────────────────────────────────────────────────────────────────────┐ │ CREWAI MULTI-AGENT ORCHESTRATION ENGINE │ │ │ │ Agent 1 ──▶ Agent 2 ──▶ Agent 3 ──▶ Agent 4 ──▶ Agent 5 │ │ Disaster Incident Survivor Medical Priority │ │ Intelligence Understanding Probability Triage Ranking │ │ │ │ │ ┌────────────────┬──────────────┘ │ │ ▼ ▼ │ │ Agent 6 Agent 7 │ │ Resource Hospital │ │ Allocation Coordination │ │ │ │ │ │ └────────┬───────┘ │ │ ▼ │ │ Agent 8 │ │ Risk Prediction │ │ │ │ │ ▼ │ │ Agent 9 │ │ Communication │ │ │ │ │ ▼ │ │ Agent 10 │ │ Command Orchestrator │ │ │ │ │ ┌───────▼───────┐ │ │ │ RESCUE PLAN │ │ │ │ JSON OUTPUT │ │ │ └───────────────┘ │ │ │ │ LLM Provider (env-configurable): │ │ ● AWS Bedrock (Claude 3 Sonnet) ● OpenAI GPT-4o │ │ ● Google Gemini Pro ● Ollama (local) ● Mock (offline) │ └──────────────────────────────────────┬─────────────────────────────────────┘ │ ┌─────────────────────────┼──────────────────────────┐ ▼ ▼ ▼ ┌────────────────────┐ ┌────────────────────────┐ ┌───────────────────────┐ │ TWILIO LAYER │ │ AWS SERVICES │ │ REAL DATASETS (DB) │ │ │ │ │ │ │ │ · SMS to field │ │ · S3 (plan storage) │ │ 547 Hospitals │ │ teams via REST │ │ · SNS (fan-out alerts) │ │ 63 Blood Banks │ │ · WhatsApp alerts │ │ · Bedrock (LLM API) │ │ 78 Fire Stations │ │ · Voice calls to │ │ · Secrets Manager │ │ 108 Ambulance Hubs │ │ hospital admin │ │ · ECR (Docker registry)│ │ 33 NDRF Units │ │ · TwiML voice │ │ · CloudWatch (logs) │ │ 47 Police Stations │ │ scripts │ │ · IAM (auth) │ │ 29 Disaster Offices │ └────────────────────┘ └────────────────────────┘ └───────────────────────┘ ## 🤖 The 10-Agent AI Pipeline ┌──────────────────────────────────────────────────────────────────────────────────┐ │ RESCUENET AI — 10-AGENT PIPELINE DETAIL │ └──────────────────────────────────────────────────────────────────────────────────┘ 📡 RAW INCIDENT REPORT (title, description, location, type) │ ▼ ╔══════════════════════════════════════════════════════════════════════════╗ ║ AGENT 1 — DISASTER INTELLIGENCE ANALYST ║ ║ Role: Classify incident type, severity, and geographic impact zone ║ ║ Input: Raw incident report ║ ║ Output: Disaster type · Severity band (1–5) · Impact radius (km) ║ ╚══════════════════════════════════════════════════════════════════════════╝ │ ↓ passes: {disaster_type, severity, impact_radius} ▼ ╔══════════════════════════════════════════════════════════════════════════╗ ║ AGENT 2 — INCIDENT COMPREHENSION SPECIALIST ║ ║ Role: Parse fragmented reports into normalized structured data ║ ║ Input: Classified incident from Agent 1 ║ ║ Output: GPS coordinates · Affected population · Infrastructure damage ║ ╚══════════════════════════════════════════════════════════════════════════╝ │ ↓ passes: {coordinates, affected_population, damage_level} ▼ ╔══════════════════════════════════════════════════════════════════════════╗ ║ AGENT 3 — SURVIVOR RISK ESTIMATION SPECIALIST ║ ║ Role: Calculate survivor probability and people needing rescue ║ ║ Input: Structured incident object, severity score ║ ║ Output: Survivor probability (0–1) · Estimated count · Time-sensitivity ║ ╚══════════════════════════════════════════════════════════════════════════╝ │ ↓ passes: {survivor_probability, estimated_survivors, time_sensitivity} ▼ ╔══════════════════════════════════════════════════════════════════════════╗ ║ AGENT 4 — MEDICAL EMERGENCY TRIAGE COORDINATOR ║ ║ Role: Assess medical needs and recommend response level ║ ║ Input: Survivor probability data, incident type ║ ║ Output: Priority level (critical/high/medium/low) · Trauma types · Resources ║ ╚══════════════════════════════════════════════════════════════════════════╝ │ ↓ passes: {medical_priority, trauma_types, resource_needs} ▼ ╔══════════════════════════════════════════════════════════════════════════╗ ║ AGENT 5 — EMERGENCY RESPONSE PRIORITIZATION OFFICER ║ ║ Role: Rank incident and determine dispatch urgency ║ ║ Input: Triage output + survivor probability + incident queue ║ ║ Output: Priority rank (P1–P5) · Urgency score · Response window ║ ╚══════════════════════════════════════════════════════════════════════════╝ │ ↓ branches into parallel tracks: ├─────────────────────────────────┐ ▼ ▼ ╔═══════════════════════════╗ ╔═══════════════════════════════════════╗ ║ AGENT 6 — RESOURCE ║ ║ AGENT 7 — HOSPITAL COORDINATION ║ ║ ALLOCATION COORDINATOR ║ ║ LIAISON & BED CAPACITY COORD. ║ ║ ║ ║ ║ ║ Optimal mix of rescue ║ ║ Best hospitals by proximity, ║ ║ teams, vehicles, equip. ║ ║ capacity & specialization ║ ║ ║ ║ ║ ║ Output: ║ ║ Output: ║ ║ · 12 ambulances ║ ║ · Gandhi Hospital (2.1km, 120 beds) ║ ║ · 2 rescue helicopters ║ ║ · Osmania Hospital (3.8km, 80 beds) ║ ║ · 4 NDRF fire teams ║ ║ · Patient routing plan ║ ╚═══════════════════════════╝ ╚═══════════════════════════════════════╝ │ │ └──────────────┬──────────────────┘ ▼ ╔══════════════════════════════════════════════════════════════════════════╗ ║ AGENT 8 — DISASTER RISK FORECASTING ANALYST ║ ║ Role: Assess secondary risks (aftershocks, flooding, fire spread) ║ ║ Input: Incident type + location + historical data ║ ║ Output: Risk evolution score · Secondary risk flags · Precautions ║ ╚══════════════════════════════════════════════════════════════════════════╝ │ ↓ passes: {risk_score, secondary_risks[], precautionary_measures[]} ▼ ╔══════════════════════════════════════════════════════════════════════════╗ ║ AGENT 9 — EMERGENCY COMMUNICATIONS OFFICER ║ ║ Role: Draft alerts for field teams, hospitals, and public ║ ║ Input: Final rescue plan, recipient types ║ ║ Output: SMS text · Hospital MCI alert · Public broadcast · TwiML ║ ║ → Triggers TWILIO dispatch automatically ║ ╚══════════════════════════════════════════════════════════════════════════╝ │ ↓ passes: {field_team_sms, hospital_alert, public_message, twiml_script} ▼ ╔══════════════════════════════════════════════════════════════════════════╗ ║ AGENT 10 — RESCUE COMMAND ORCHESTRATOR ║ ║ Role: Synthesize ALL 9 agent outputs into one rescue plan ║ ║ Input: Context from ALL prior agents (t1 through t9) ║ ║ Output: Final RescuePlan JSON — the definitive field command ║ ╚══════════════════════════════════════════════════════════════════════════╝ │ ▼ ┌──────────────────────────────────────────────────────────────┐ │ RESCUE PLAN JSON OUTPUT │ │ { │ │ "priority": "P1", │ │ "severity": 5, │ │ "affected_area": "Begumpet, Hyderabad", │ │ "estimated_survivors": 4200, │ │ "survivor_probability": 0.72, │ │ "medical_priority": "critical", │ │ "recommended_hospital": "Gandhi Hospital", │ │ "recommended_resources": [ │ │ {"type": "ambulance", "count": 12, "eta_minutes": 8}, │ │ {"type": "helicopter", "count": 2, "eta_minutes": 15}, │ │ {"type": "ndrf_team", "count": 4, "eta_minutes": 22} │ │ ], │ │ "risk_warnings": ["Flash flood escalation in 6 hrs"], │ │ "alert_actions": { │ │ "field_team": "PRIORITY P1: Deploy immediately...", │ │ "hospital": "MCI ALERT: Expect 350 casualties...", │ │ "public": "EVACUATION ORDER: Begumpet zone..." │ │ } │ │ } │ └──────────────────────────────────────────────────────────────┘ │ │ │ ▼ ▼ ▼ 📱 Twilio SMS 📞 Voice Call 🖥️ Dashboard to field teams to hospitals real-time update ### Agent Context Chaining (How Agents Talk to Each Other) | Task | Agent | Receives Context From | Key Value Add | |---|---|---|---| | Task 1 | Disaster Intelligence | _(Raw report only)_ | Sets disaster classification for all downstream agents | | Task 2 | Incident Understanding | Task 1 | Converts vague text into GPS + population numbers | | Task 3 | Survivor Probability | Task 2 | Applies statistical model → survivor count | | Task 4 | Medical Triage | Task 3 | Translates survivor count into medical resource needs | | Task 5 | Priority Agent | Tasks 1–4 | Full picture → accurate P1–P5 ranking | | Task 6 | Resource Allocation | Tasks 4, 5 | Medical needs + priority → correct unit mix | | Task 7 | Hospital Coordination | Tasks 3, 4, 5 | Routes casualties to real Telangana hospitals | | Task 8 | Risk Prediction | Tasks 1, 2 | Secondary hazard forecasting from incident type + location | | Task 9 | Communication | Tasks 5, 6, 7, 8 | Accurate, data-backed alert messages for all recipients | | Task 10 | Command Orchestrator | Tasks **1–9 ALL** | Synthesizes, not invents — the definitive rescue command | ## ☁️ AWS Infrastructure ┌───────────────────────────────────────────────────────────────────────────────────┐ │ ☁️ AWS CLOUD INFRASTRUCTURE │ └───────────────────────────────────────────────────────────────────────────────────┘ 👤 Emergency Commanders · Field Teams · Hospital Admins │ HTTPS ▼ ┌────────────────────────────────────┐ │ AWS AMPLIFY (CDN) │ ← Next.js 14, global distribution, │ Next.js 14 Frontend │ CI/CD from GitHub, SSL termination └─────────────────┬──────────────────┘ │ REST API Calls ▼ ┌────────────────────────────────────┐ ┌──────────────────────────────────┐ │ AMAZON EC2 (Docker) │────▶│ AMAZON ECR │ │ FastAPI Backend │ │ Docker Registry │ │ Pydantic · CrewAI Orchestrator │ │ rescuenet-backend:latest │ └──────────────────┬─────────────────┘ └──────────────────────────────────┘ │ ▲ ┌───────────┼──────────────────────┐ │ ▼ ▼ ▼ ▼ ┌─────┴────────────────┐ ┌──────────┐ ┌──────────┐ ┌────────┐ ┌──────────┐ │ AWS IAM │ │ AWS │ │ AMAZON │ │ AMAZON │ │ AWS │ │ Role-Based Auth │ │ BEDROCK │ │ SNS │ │ S3 │ │ SECRETS │ │ ECR Token Generator │ │ │ │ │ │ │ │ MANAGER │ └──────────────────────┘ │ Claude 3 │ │ Fan-out │ │ Plan │ │ │ │ Sonnet │ │ Alerts │ │ Store │ │ API Keys │ ┌──────────────────────┐ │ LLM API │ │ Multi-ch.│ │ Audit │ │ DB URIs │ │ AMAZON CLOUDWATCH │ └──────────┘ └────┬─────┘ │ Logs │ │ Twilio │ │ Agent Step Logs │ │ └────────┘ └──────────┘ │ Metrics · Alarms │ ┌──────────┴─────┐ └──────────────────────┘ ▼ ▼ ┌────────────┐ ┌────────────┐ │ TWILIO │ │ AMAZON SES │ ┌─────────────────────────────┐ │ SMS/Voice │ │ Email │ │ NEON SERVERLESS POSTGRES │ │ WhatsApp │ │ Reports │ │ Disaster Asset Database │ └────────────┘ └────────────┘ │ Incidents · Rescue Plans │ │ Agent Decisions · Alerts │ └─────────────────────────────┘ ### AWS Services Breakdown | Service | Role | Current Status | |---|---|---| | **AWS Amplify** | Frontend CDN — Next.js build, SSL, global distribution | ✅ Configured (`amplify.yml`) | | **Amazon EC2** | FastAPI backend compute host | ✅ Active (`apprunner-config.json`) | | **Amazon ECR** | Private Docker registry for `rescuenet-backend:latest` | ✅ Integrated | | **AWS IAM** | Role-based access, short-lived ECR tokens, no hardcoded credentials | ✅ Active | | **AWS Bedrock** | Primary production LLM (Claude 3 Sonnet) | ⚙️ Configured (env-based) | | **Amazon SNS** | Fan-out alert dispatch to Twilio, SES, and SMS endpoints | ⚙️ ARN configured | | **Amazon S3** | Rescue plan JSON storage, audit log archiving | ⚙️ Bucket configured | | **AWS Secrets Manager** | Secure storage for DB URIs, API keys, Twilio tokens | 🔵 Planned (v0.3) | | **Amazon CloudWatch** | Agent step logging, performance metrics, alarm triggers | 🔵 Planned (v0.3) | ### Deployment Script The backend is deployed via `scripts/deploy_to_ec2.ps1`: # 1. Authenticate with AWS IAM → get ECR login token # 2. Upload execution script to EC2 via SSH # 3. Auto-install Docker + AWS CLI if missing # 4. Pull rescuenet-backend:latest from ECR # 5. Run container with all env vars (DB URI, CORS, LLM keys) # Full deployment checklist: # ✅ Push Docker image to Amazon ECR # ✅ Provision EC2 instance, open port 8000 # ✅ Provision Neon PostgreSQL instance # ✅ Execute deploy_to_ec2.ps1 # ✅ Connect Amplify to GitHub repo (auto-deploy on push) # ✅ Set NEXT_PUBLIC_API_URL in Amplify env vars # ✅ Verify: GET http://:8000/health → {"status":"ok"} ## 📱 Twilio Automation Layer RescueNet AI uses **Twilio** as the final-mile alert dispatch engine, triggered automatically by Agent 9 (Communication Agent) after the rescue plan is assembled. ┌─────────────────────────────────────────────────────────────────────────┐ │ TWILIO AUTOMATION PIPELINE │ └─────────────────────────────────────────────────────────────────────────┘ Agent 9 (Communication Agent) drafts 3 alert types: ┌───────────────────────────────────────────────────────────────────────┐ │ 📱 SMS → Field Rescue Teams │ │ ─────────────────────────────────────────────────────────────────── │ │ "PRIORITY P1 — BEGUMPET FLOOD: Deploy 12 ambulances + 2 choppers │ │ immediately. Survivor est: 4,200. ETA to site: 8 min. │ │ Hospital routing: GANDHI (primary) | OSMANIA (overflow). │ │ WARNING: Flash flood escalation expected in 6 hours." │ │ │ │ Sent via: POST https://api.twilio.com/2010-04-01/Accounts/ │ │ {SID}/Messages.json │ └───────────────────────────────────────────────────────────────────────┘ ┌───────────────────────────────────────────────────────────────────────┐ │ 📞 VOICE CALL → Hospital MCI Director │ │ ─────────────────────────────────────────────────────────────────── │ │ TwiML Script: │ │ "This is RescueNet AI Emergency Alert. A P1 Mass Casualty │ │ Incident has been declared in Begumpet. Gandhi Hospital: │ │ Prepare to receive 350 casualties. Trauma and burns primary. │ │ ETA first wave: 22 minutes. Activate MCI protocol now." │ │ │ │ Sent via: Twilio Programmable Voice + TwiML │ └───────────────────────────────────────────────────────────────────────┘ ┌───────────────────────────────────────────────────────────────────────┐ │ 💬 WHATSAPP → Command Center Supervisors │ │ ─────────────────────────────────────────────────────────────────── │ │ Rich message with incident summary, resource list, hospital routing │ │ Sent via Twilio WhatsApp Sandbox / Business API │ └───────────────────────────────────────────────────────────────────────┘ ### Twilio Configuration # .env variables required for live Twilio dispatch: TWILIO_ACCOUNT_SID=AC... # Your Twilio Account SID TWILIO_AUTH_TOKEN=... # Your Twilio Auth Token TWILIO_WHATSAPP_FROM=whatsapp:+14155238886 # WhatsApp Sandbox number TWILIO_VOICE_FROM=+1... # Voice-capable phone number PUBLIC_URL=https://your-api-url.com # For TwiML webhook callbacks SNS_ALERT_TOPIC_ARN=arn:aws:sns:... # AWS SNS for fan-out ### Full Automation Trigger Flow POST /api/v1/agents/execute { incident_id } → Agent 9 output: { field_team_sms, hospital_twiml, whatsapp_msg } → communication_service.dispatch_alerts() ├── Twilio SMS API → field team mobile numbers ├── Twilio Voice API → hospital MCI director number ├── Twilio WhatsApp → command center supervisor └── AWS SNS publish → fan-out to additional subscribers → alert_logs written to DB (channel, recipient, status, timestamp) ## 🗺️ Real Research Datasets — Telangana RescueNet AI is powered by **7 verified emergency asset datasets** compiled from official Telangana government and national emergency management sources. These are not mock data — these are real facility records. ┌─────────────────────────────────────────────────────────────────────────┐ │ TELANGANA EMERGENCY ASSET DATABASE │ │ (~905 verified records) │ └─────────────────────────────────────────────────────────────────────────┘ | Dataset | Records | Source | Coverage | |---|---|---|---| | 🏥 **Hospitals** | 547 | Telangana Health Dept. | Gandhi, Osmania, MGM Warangal, NIMS, + district hospitals | | 🩸 **Blood Banks** | 63 | Indian Red Cross / TNBCS | All 33 districts | | 🚒 **Fire Stations** | 78 | Telangana State Disaster Response Force | Fire tenders + rescue units per district | | 🚑 **Ambulance Services** | 108 | GVK EMRI 108 / TSEMS | Dispatch hubs across Telangana | | 🪖 **NDRF Units** | 33 | National Disaster Response Force | Battalions + rapid deployment teams | | 👮 **Police Stations** | 47 | Telangana Police | Control rooms + emergency response units | | 🏢 **Disaster Mgmt. Offices** | 29 | TSSDMA | State + district disaster management authorities | ### Dataset Files in Repository rescuenet-ai/ ├── Telangana Hospitals Dataset Compilation.pdf # 547 hospitals, full metadata ├── Telangana Blood Banks Dataset.pdf # 63 blood banks across districts ├── Telangana – Fire Stations Dataset.pdf # 78 fire stations + equipment ├── telangana_ambulance_services.pdf # 108 GVK EMRI dispatch hubs ├── telangana_ndrf_units.pdf # 33 NDRF units + battalion codes ├── telangana_police_stations.pdf # 47 stations + control rooms ├── telangana_disaster_management_offices.pdf # 29 TSSDMA offices ├── emergency_assets_master.csv # Consolidated master dataset └── emergency_assets_clean.csv # Cleaned + normalized for DB ### Key Facilities Used in Agent Routing | Facility | Type | District | Role in Rescue Plan | |---|---|---|---| | **Gandhi Hospital** | Tertiary (Govt.) | Hyderabad | Primary trauma + burn center | | **Osmania General Hospital** | Tertiary (Govt.) | Hyderabad | Secondary overflow, general trauma | | **NIMS** | Super-specialty | Hyderabad | Neurology + critical care overflow | | **MGM Hospital Warangal** | Tertiary (Govt.) | Warangal | North Telangana primary response | | **GVK EMRI 108 Hub (Begumpet)** | Ambulance dispatch | Hyderabad | Fastest 108 deployment point | | **Hyderabad NDRF Battalion** | Rapid rescue team | Hyderabad | Flood + collapse rescue operations | ## 🛠️ Tech Stack | Layer | Technology | Purpose | |---|---|---| | **Frontend** | Next.js 14 (App Router), TypeScript, Tailwind CSS | Dashboard, incident form, map view | | **Backend** | FastAPI, Python 3.11, Pydantic v2, Uvicorn | REST API, validation, orchestration | | **Agent Engine** | CrewAI (multi-agent framework) | 10-agent sequential pipeline | | **LLM (Primary)** | AWS Bedrock — Claude 3 Sonnet | Production inference | | **LLM (Alt.)** | OpenAI GPT-4o / Google Gemini / Ollama | Provider-agnostic fallback | | **Database** | Neon Serverless PostgreSQL | Incident, rescue plan, asset data | | **Frontend Hosting** | AWS Amplify | CDN, CI/CD, SSL | | **Backend Hosting** | Amazon EC2 + Docker | Dedicated compute | | **Container Registry** | Amazon ECR | Private Docker image hosting | | **LLM Infrastructure** | AWS Bedrock | Managed LLM API | | **Alert Fan-out** | Amazon SNS | Multi-channel alert broadcast | | **SMS / Voice / WhatsApp** | Twilio | Real-time emergency dispatch | | **Auth & Security** | AWS IAM, Secrets Manager | Credential management | | **Maps** | Google Maps API | Incident and resource geolocation | | **Observability** | Amazon CloudWatch | Logs, metrics, alarms | ## 🚀 Quick Start ### Prerequisites - Python 3.11+ - Node.js 18+ - Docker & Docker Compose - AWS CLI (for deployment) - LLM API key — OR use `LLM_PROVIDER=mock` for full offline demo

🛠️ 1. Clone and Configure

git clone https://github.com/Vikram30069/RescueNet-AI.git cd RescueNet-AI cp .env.example .env # Edit .env: set LLM_PROVIDER, DATABASE_URL, and optionally TWILIO_* keys

🐳 2. Run with Docker (Fastest)

docker-compose up --build | Service | URL | |---|---| | 🖥️ Frontend Dashboard | http://localhost:3000 | | 🔌 Backend API | http://localhost:8000 | | 📋 Swagger Docs | http://localhost:8000/docs | | 💓 Health Check | http://localhost:8000/health |

💻 3. Run Manually (Local Dev)

**Backend:** cd backend python -m venv venv venv\Scripts\activate # Windows source venv/bin/activate # Mac/Linux pip install -r requirements.txt uvicorn app.main:app --reload --port 8000 **Frontend:** cd frontend npm install npm run dev **Run Agent Pipeline (standalone test):** python scripts/run_agents.py **Trigger full demo scenario:** python trigger_demo.py

☁️ 4. Deploy to AWS

# Push Docker image to ECR: aws ecr get-login-password --region us-east-1 | docker login --username AWS --password-stdin docker build -t rescuenet-backend . docker tag rescuenet-backend:latest /rescuenet-backend:latest docker push /rescuenet-backend:latest # Deploy to EC2: .\scripts\deploy_to_ec2.ps1 # Frontend: connect AWS Amplify to this GitHub repo # Set env var: NEXT_PUBLIC_API_URL=http://:8000

## 📂 Repository Structure RescueNet-AI/ ├── 🖥️ frontend/ → Next.js 14 App Router (TypeScript, Tailwind) ├── 🔌 backend/ → FastAPI Python server │ └── app/ │ ├── routers/ → /incidents /agents /hospitals /resources │ ├── services/ → Business logic + agent triggering │ ├── schemas/ → Pydantic request/response models │ └── db/ → PostgreSQL repository layer ├── 🤖 agents/ → CrewAI multi-agent orchestration │ ├── definitions/ → 10 agent definition files │ ├── tasks/ → Task chain with context passing │ ├── config/ → LLM provider config (env-based) │ └── orchestrator.py → Pipeline runner + rescue plan assembler ├── 🗄️ database/ → SQL schema files ├── 🌱 seed/ → Demo seed data (SQL + Python) ├── 📜 scripts/ → Developer utilities + EC2 deploy script ├── 📊 docs/ → Diagrams, architecture references ├── 🗺️ [Telangana Datasets] → 7 real emergency asset datasets (PDF + CSV) ├── docker-compose.yml ├── amplify.yml → AWS Amplify CI/CD config ├── apprunner-config.json → AWS App Runner config └── .env.example → All environment variable templates ## 🔌 API Reference ### Core Endpoints | Method | Endpoint | Description | |---|---|---| | `GET` | `/health` | System health check | | `POST` | `/api/v1/incidents` | Submit a new incident | | `GET` | `/api/v1/incidents` | List all incidents | | `GET` | `/api/v1/incidents/{id}` | Get incident by ID | | `POST` | `/api/v1/agents/execute` | Trigger 10-agent pipeline | | `GET` | `/api/v1/rescue-plan/{incident_id}` | Retrieve rescue plan | | `GET` | `/api/v1/hospitals` | List available hospitals | | `GET` | `/api/v1/resources` | List available resources | ### Trigger the Full AI Pipeline # Step 1: Submit an incident curl -X POST http://localhost:8000/api/v1/incidents \ -H "Content-Type: application/json" \ -d '{ "title": "Begumpet Flash Flood", "description": "350mm rainfall in 12 hours. Thousands stranded.", "location": "Begumpet, Hyderabad, Telangana", "disaster_type": "flood", "severity": 5 }' # Step 2: Execute the 10-agent pipeline curl -X POST http://localhost:8000/api/v1/agents/execute \ -H "Content-Type: application/json" \ -d '{"incident_id": "YOUR_INCIDENT_ID"}' # Response: Full RescuePlan JSON + 10 agent decision logs ### Sample Rescue Plan Response { "priority": "P1", "severity": 5, "affected_area": "Begumpet, Hyderabad", "estimated_survivors": 4200, "survivor_probability": 0.72, "medical_priority": "critical", "dispatch_urgency": "immediate", "recommended_hospital": "Gandhi Hospital", "recommended_resources": [ {"type": "ambulance", "count": 12, "eta_minutes": 8}, {"type": "helicopter", "count": 2, "eta_minutes": 15}, {"type": "ndrf_team", "count": 4, "eta_minutes": 22} ], "hospitals": [ {"name": "Gandhi Hospital", "distance_km": 2.1, "available_beds": 120, "patient_routing": 250}, {"name": "Osmania Hospital", "distance_km": 3.8, "available_beds": 80, "patient_routing": 100} ], "risk_warnings": ["Flash flood escalation expected in 6 hours", "Structural collapse risk in low-lying zones"], "alert_actions": { "field_team": "PRIORITY P1: Deploy 12 ambulances + 2 helicopters to Begumpet immediately...", "hospital": "MCI ALERT: Gandhi Hospital — expect 350 casualties, trauma + drowning cases...", "public": "EVACUATION ORDER: Begumpet zone — move to elevated ground immediately..." }, "agent_decisions": [ {"step": 1, "agent": "disaster_intelligence", "output": {...}}, {"step": 2, "agent": "incident_understanding", "output": {...}}, "... 10 decisions total ..." ] } ## ⚙️ Configuration ### LLM Provider Selection # .env — choose ONE provider: LLM_PROVIDER=mock # Offline, instant, deterministic (demo/dev) LLM_PROVIDER=openai # OpenAI GPT-4o (requires OPENAI_API_KEY) LLM_PROVIDER=gemini # Google Gemini Pro (requires GEMINI_API_KEY) LLM_PROVIDER=ollama # Local Ollama (requires ollama running on :11434) LLM_PROVIDER=litellm # LiteLLM proxy (multi-provider) | Provider | Avg Pipeline Time | Cost | Best For | |---|---|---|---| | `mock` | < 1 second | Free | Development, demos, testing | | `ollama` | 5–30 seconds | Free (GPU required) | Privacy-first local runs | | `gemini` | 10–40 seconds | Pay-per-use | Google ecosystem | | `openai` | 10–45 seconds | Pay-per-use | Highest quality outputs | | `bedrock` | 10–30 seconds | AWS pricing | Production deployment | ## 📚 Key Documents | Document | Purpose | |---|---| | 📋 [AGENTS.md](./AGENTS.md) | All 10 CrewAI agent specs (roles, inputs, outputs) | | 🏗️ [ARCHITECTURE.md](./ARCHITECTURE.md) | System design, component boundaries | | ☁️ [AWS_DEPLOYMENT.md](./AWS_DEPLOYMENT.md) | Full AWS cloud deployment guide | | 🔄 [WORKFLOWS.md](./WORKFLOWS.md) | End-to-end incident workflow with Mermaid diagrams | | 🔗 [AGENT_DEPENDENCY_GRAPH.md](./AGENT_DEPENDENCY_GRAPH.md) | Agent context chaining visualization | | 📊 [TASK_CHAINING_REPORT.md](./TASK_CHAINING_REPORT.md) | Technical deep-dive: before/after context passing | | 🗄️ [DATABASE_SCHEMA.md](./DATABASE_SCHEMA.md) | PostgreSQL table definitions | | 🔌 [API_SPEC.md](./API_SPEC.md) | REST endpoint contracts + examples | | 🔍 [PROJECT_AUDIT.md](./PROJECT_AUDIT.md) | Full codebase audit + gap analysis | | 🚑 [RECOVERY_GUIDE.md](./RECOVERY_GUIDE.md) | How to restore and recover the system | ## 🗺️ Roadmap | Phase | Version | Feature | |---|---|---| | ✅ | **v0.1** | 10 CrewAI agents + mock mode + full pipeline + Neon DB + AWS deploy | | 🔵 | **v0.2** | Live Twilio SMS + Amazon Connect voice calls (wired, not mock) | | 🔵 | **v0.3** | AWS Bedrock Claude production LLM + Secrets Manager + CloudWatch | | 🔵 | **v0.4** | Real-time IoT/sensor incident stream + WebSocket live agent progress | | 🔵 | **v0.5** | ML-based disaster risk forecasting using historical Telangana data | | 🔵 | **v0.6** | Full interactive map with incident pins + resource/hospital overlays | | 🔵 | **v1.0** | Production-grade AWS ECS Fargate + RDS + ALB + WAF deployment | | 🔮 | **v2.0** | Expand beyond Telangana → national NDMA integration | ## 👥 User Personas | Persona | Role | Primary Interface | |---|---|---| | **Emergency Commander** | Reviews AI rescue plan, approves dispatch | Dashboard + Rescue Plan Viewer | | **Field Coordinator** | Receives SMS, coordinates on-ground teams | Twilio SMS + WhatsApp | | **Hospital Administrator** | Accepts incoming casualty alerts, updates bed count | Voice Call + Dashboard | | **System Operator** | Manages RescueNet platform, reviews audit logs | Admin Dashboard + API | ## 💓 Health Check GET http://localhost:8000/health → { "status": "ok", "version": "0.1.0", "agents": 10, "mode": "mock" } ## 📜 License MIT — Built for India's disaster response ecosystem.

**Built with ❤️ for Telangana · Ready to scale for India** _RescueNet AI — When every second counts, AI coordinates._

标签：AV绕过, AWS, CrewAI, DPI, FastAPI, PyRIT, 人工智能, 多智能体系统, 测试用例, 灾害应急管理, 用户模式Hook绕过, 请求拦截, 逆向工具