🛡️ DRISHTi: Real-Time Disaster Early-Warning & Resource Coordination Platform

An end-to-end disaster-response decision-support platform designed for Rourkela, Odisha, India. The platform converts unstructured citizen incident reports into structured AI intelligence, computes a transparent 0–100 priority score, evaluates Haversine distance matrix, executes SciPy Hungarian bipartite resource matching, streams real-time updates via WebSockets, and provides interactive GIS command dashboards.

---

🌐 Official Live Links (100% Active & Padlock Verified)

Resource	Live Link	Security & Details
🔒 Official HTTPS Web App	https://disasterresponse.click	🟢 SECURE (100% Trusted Let's Encrypt SSL Padlock)
🌐 WWW Subdomain (HTTPS)	https://www.disasterresponse.click	🟢 SECURE (Trusted TLS 1.3 Encryption)
📑 Interactive Swagger API Docs	https://disasterresponse.click/docs	🟢 Live OpenAPI REST API Explorer
🔒 Cloudflare Edge Mirror	https://therefore-pointing-downtown-save.trycloudflare.com	🟢 SECURE (Cloudflare CDN Edge)
🌐 Direct Cloud IP	http://13.204.160.135	🟢 Amazon EC2 (Mumbai ap-south-1)

---

🏗️ System Architecture

flowchart TD
    subgraph Clients["User Interfaces"]
        CitizenUI["Citizen React Frontend<br/>(Neo-Brutalist DRISHTi UI)"]
        DashboardUI["Authority Command Dashboard<br/>(Tactical Map & Queue)"]
    end

    subgraph Nginx["Nginx Reverse Proxy (Port 80 / 443)"]
        Proxy["SSL Termination (Let's Encrypt)<br/>& WebSocket Upgrades"]
    end

    subgraph Backend["FastAPI Backend Engine (Port 8000)"]
        API["FastAPI REST & WebSocket Controllers<br/>(/api/v1)"]
        GeminiService["Gemini AI & Fallback NLP Extractor"]
        ScoringEngine["Transparent Priority Engine (0-100)"]
        SciPyOptimizer["SciPy Allocation Engine<br/>(linear_sum_assignment)"]
        HaversineMath["Haversine Distance Calculator"]
        WSManager["WebSocket Event Broadcaster"]
    end

    subgraph Database["Relational Store (Port 5432)"]
        DB[(PostgreSQL 15 / Persistent Volume)]
    end

    CitizenUI -->|HTTPS / WSS| Proxy
    DashboardUI -->|HTTPS / WSS| Proxy
    Proxy --> API
    
    API --> GeminiService
    API --> ScoringEngine
    API --> SciPyOptimizer
    SciPyOptimizer --> HaversineMath
    ScoringEngine --> HaversineMath
    
    API --> DB
    API --> WSManager
    WSManager -.->|Real-time WS Events| Proxy

Loading

---

⚡ Quick Start with Docker Compose

To start all services locally with Docker Compose:

docker compose up --build

Access URLs

• Citizen React Portal: http://localhost:5173
• FastAPI Backend & Interactive Swagger Docs: http://localhost:8000/docs
• Streamlit Authority Command Dashboard: http://localhost:8501

---

🚀 Production AWS Deployment (<$10/Month)

The platform is deployed on AWS EC2 (Mumbai ap-south-1) with persistent PostgreSQL 15, FastAPI, React 18, and Nginx.

# Provision Infrastructure
cd terraform
terraform init
terraform apply -auto-approve

---

🔑 Environment Variables

Variable	Default Value	Description
GEMINI_API_KEY	(Optional)	Google Gemini API key. If omitted or API fails, heuristic fallback parser is active.
GEMINI_MODEL	gemini-2.5-flash	Gemini model version for structured JSON extraction.
DATABASE_URL	postgresql://...	Database connection URI (PostgreSQL 15 in production, SQLite in dev).
API_BASE_URL	http://localhost:8000/api/v1	Backend API URL for dashboard & frontend communication.
VITE_API_BASE_URL	/api/v1	React frontend Vite backend endpoint.

---

📊 Priority Scoring Formula & Deterministic Engine

The priority score $P \in [0, 100]$ is computed deterministically using five weighted normalized components ($C_i \in [0, 100]$):

$$P = 0.35 \times C_{\text{severity}} + 0.25 \times C_{\text{people}} + 0.15 \times C_{\text{facility}} + 0.15 \times C_{\text{resource}} + 0.10 \times C_{\text{time}}$$

Factor Mappings

1. Severity ($C_{\text{severity}}$): LOW = 25, MEDIUM = 50, HIGH = 80, CRITICAL = 100.
   • Vulnerability Promotion Rule: If vulnerable individuals (children/elderly/disabled) are present and severity < HIGH, severity is promoted by one enum step.
2. People Affected ($C_{\text{people}}$): 0 $\rightarrow$ 0; 1 $\rightarrow$ 20; 2–5 $\rightarrow$ 40; 6–10 $\rightarrow$ 65; 11–25 $\rightarrow$ 85; 26+ $\rightarrow$ 100 (+10 pts if vulnerable people present, capped at 100).
3. Facility Proximity ($C_{\text{facility}}$): $\max(0, 100 - (\text{dist}_{\text{km}} / 20) \times 100)$ to nearest active critical facility.
4. Resource Availability ($C_{\text{resource}}$): 100 (<2 km), 85 (<5 km), 65 (<10 km), 40 (<20 km), 20 (>20 km), 0 (none compatible).
5. Time Elapsed ($C_{\text{time}}$): $\min(100, (\text{elapsed_minutes} / 180) \times 100)$.

Priority Classification

• 70–100: HIGH PRIORITY (Red)
• 40–69: MEDIUM PRIORITY (Yellow/Amber)
• 0–39: LOW PRIORITY (Green)

---

🤖 SciPy Resource Allocation Engine

Global multi-incident resource allocation uses scipy.optimize.linear_sum_assignment over an incident-by-resource cost matrix:

$$\text{Cost}_{i, j} = \text{Distance}_{\text{km}} + 0.75 \times (100 - P_i) + \text{Penalty}_{\text{capability}} + \text{Penalty}_{\text{capacity}}$$

• Capability Penalty: 0 for exact match, 35 for valid fallback, 1_000_000 for infeasible pairs.
• Scarcity Behavior: High-priority incidents ($P_i \ge 70$) are strongly favored for nearby resources over slightly closer low-priority incidents.
• Transactional Confirmation: Optimization produces previews; dispatching calls /api/v1/assignments/confirm to update resource status AVAILABLE $\rightarrow$ BUSY and incident status REPORTED $\rightarrow$ ASSIGNED.

---

🎬 3-Minute Hackathon Demo Script

1. Trigger Alert: Click ⚡ TRIGGER SYNTHETIC ALERT for Flood in Rourkela.
2. Verify Alert: Observe live alert ticker in Citizen Portal.
3. Submit Citizen Report: Click sample chip: "Water has entered several houses and 8 people are trapped. Two of them are elderly."
4. Inspect AI & Score: Observe structured extraction (8 people, vulnerable true, High severity) and HIGH priority classification score ($\ge 70$).
5. Inspect 3D Map: View incident column on tactical map.
6. Run SciPy Optimizer: Click ⚙️ RUN SCIPY OPTIMIZE ALGORITHM, inspect match recommendations, distance, ETA, and cost.
7. Confirm Dispatch: Click CONFIRM DISPATCH. Observe resource change AVAILABLE $\rightarrow$ BUSY and incident change to ASSIGNED.
8. Complete Lifecycle: Advance assignment to COMPLETED. Verify resource returns to AVAILABLE and incident becomes RESOLVED.

---

🧪 Testing & CI/CD

Run backend unit & integration tests:

cd backend
PYTHONPATH=. pytest app/tests/ -v

Run frontend build check:

cd frontend
npm run build