Huzair43/sahel-security-analysis

Data-driven analysis of security incidents in the Sahel (Burkina Faso, Mali, Niger) over 10+ years, combining conflict data with economic indicators to uncover patterns, trends, and geopolitical insights. Interactive dashboard included. # Sahel Security Analysis An open-source intelligence platform analyzing armed conflict dynamics and their economic consequences in Burkina Faso, Mali, and Niger, built on ACLED data, Bayesian modeling, and a RAG-powered AI analyst. **Live demo:** sahel-security-analysis.onrender.com First load takes 30 to 60 seconds: the free Render instance spins down when idle. --- *23,156 ACLED incidents rendered with Plotly WebGL. Marker size encodes fatalities, colour encodes event type, filtering runs entirely in the browser.* --- ## What it does - Visualizes 23,000+ conflict incidents on an interactive WebGL map with real-time client-side filtering - Tracks monthly trends, detects anomalies (Z-score), and projects a 6-month forecast via linear regression - Identifies geographic hotspots and most active armed groups - Quantifies the conflict-inflation relationship through a Bayesian hierarchical model (PyMC, ArviZ) - Exposes an AI analyst powered by a RAG pipeline: answers are constrained to context retrieved from the ACLED dataset *Key metrics, distribution by country and by event type, monthly trend.* *Rolling averages, fitted linear trend and a 6-month baseline projection.* *Questions answered from context retrieved in the ACLED DataFrame with Pandas, then passed to Llama 3.1 through Groq.* --- ## Versions | Branch | Stack | Status | |---|---|---| | `main` | Flask, Plotly WebGL, RAG chatbot, Render | Live | | `streamlit` | Streamlit, Folium | Preserved | The project was originally built with Streamlit. It was migrated to Flask to support a production deployment with a RAG chatbot, WebGL map performance at 23k points, and full control over the frontend design. --- ## Architecture ``` flask_app/ ├── routes/ # Page blueprints (overview, map, trends, hotspots, bayesian, about) ├── api/ # JSON endpoints (map points, chart data, chat) ├── chat/ │ ├── rag.py # Intent detection + Pandas retrieval + prompt builder │ ├── …