TigerSafe — Campus Dispatch Copilot

A safety-aware routing system for Mizzou students. TigerSafe combines real crime data, OpenStreetMap walking networks, and an AI-powered chat assistant to help students find safe routes across campus.

Features

• Safety-weighted routing — Finds routes that balance distance with crime density, emergency phone proximity, and time-of-day risk
• Interactive crime heatmap — Folium-based map with crime density visualization, emergency phone markers, and building overlays
• AI chat assistant — Claude-powered "TigerSafe" agent that answers safety questions, compares routes, and explains risk factors
• Temporal risk analysis — Risk scores adjust based on time of day (late night vs. daytime)
• Explainable risk scores — Every recommendation comes with a breakdown of contributing factors

Architecture

┌──────────────┐     ┌──────────────────┐     ┌──────────────┐
│  Streamlit   │────▶│  Claude Agent    │────▶│   Safety     │
│  Frontend    │     │  (tool-use API)  │     │   Engine     │
│  + Folium    │     │  5 tools         │     │  + OSMnx     │
└──────────────┘     └──────────────────┘     └──────────────┘
       │                                            │
       └──────────── Shared Data Layer ─────────────┘
                    (GeoJSON + GraphML)

Modules:

File	Purpose
app.py	Streamlit UI with Folium map and chat interface
agent.py	Claude-powered AI agent with 5 safety tools
safety.py	Risk scoring engine, safety-weighted routing
ingest.py	Data ingestion from CPD, MU ArcGIS, and OSM
seed_data.py	Sample data generator for demo/offline use

Quick Start

1. Install dependencies

python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Generate seed data

This downloads the OSM walking network (requires internet) and generates sample crime/infrastructure data:

python seed_data.py

3. Set your API key

The AI chat requires an Anthropic API key:

export ANTHROPIC_API_KEY="sk-ant-..."

Or enter it in the sidebar when the app launches.

4. Run the app

streamlit run app.py

5. In the app

1. Click "Load / Refresh Data" in the sidebar to load crime and infrastructure data
2. Click "Prepare Routing Graph" to build the safety-weighted network (takes a moment)
3. Start chatting! Try: "What's the safest route from Ellis Library to the Rec Center?"

Example Queries

• "What's the safest route from Jesse Hall to Engineering?"
• "Are there any recent incidents near Memorial Union?"
• "What safety factors should I be aware of near the Columns at night?"
• "Is the shuttle running right now?"
• "Compare the safest and fastest routes from the Student Center to Hearnes Center"

Data Sources

Source	Type	Access
OpenStreetMap (osmnx)	Walking network	API (free)
CPD Transparency Portal	Crime statistics	ArcGIS REST API
MUPD Daily Crime Log	Campus incidents	HTML scraping
MU Campus Map	Buildings, emergency phones	ArcGIS REST API
Tiger Line GTFS	Shuttle schedules	Static feed

The app falls back to realistic sample data if live APIs are unavailable.

Risk Score Algorithm

Each route receives a risk score (0-100) based on:

• Base crime score: 10 points per nearby incident
• Violent crime bonus: +15 for violent offenses
• Time-of-day multiplier: 2x for 10pm-6am, 0.3x for daytime
• Emergency phone proximity: Reduces score near blue-light phones
• Length normalization: Score is per-100m of route

Score	Level	Color
0-20	Very Safe	Green
21-40	Safe	Blue
41-60	Moderate Risk	Orange
61-80	Higher Risk	Red
81-100	High Risk	Dark Red

Project Structure

.
├── app.py              # Streamlit application
├── agent.py            # Claude AI agent with tools
├── safety.py           # Safety scoring & routing engine
├── ingest.py           # Data ingestion pipelines
├── seed_data.py        # Sample data generator
├── requirements.txt    # Python dependencies
├── data/               # Generated data files (gitignored)
│   ├── columbia_walk.graphml
│   ├── cpd_crimes.geojson
│   ├── emergency_phones.geojson
│   └── buildings.geojson
├── design.md           # Architectural design document
└── data_analysis.md    # Data source analysis