Much1r1/project_kijiji

Predictive Graph-Neural Network Frameworks for Optimizing Digital Public Infrastructure (DPI) in Sub-Saharan Africa # Project Kijiji 🌍 ### Distributed Network Observability & Predictive Modeling Platform > *"How can Graph Neural Networks predict optimal peering points to mitigate routing detours in Sub-Saharan digital infrastructure?"* A research-grade platform that maps, detects, and predicts African Internet routing inefficiencies — quantifying the latency and economic cost of traffic that unnecessarily routes through European hubs before reaching its destination. Built as a DAAD scholarship research project targeting **SDG 9.4** (resilient infrastructure). --- ## Table of Contents 1. The Research Problem 2. Architecture Overview 3. Module Breakdown 4. Tech Stack 5. Data Sources 6. Getting Started 7. File Structure 8. Research Pillars 9. Roadmap --- ## The Research Problem A packet sent from Nairobi to Lagos — two African cities ~4,000km apart — often routes through Frankfurt or London before arriving. This "trombone routing" pattern is a legacy of colonial-era infrastructure and the lack of direct IXP (Internet Exchange Point) peering agreements between African ASes (Autonomous Systems). **The cost is measurable:** - Added latency: 100–200ms per round trip - Economic waste: transit fees paid to European carriers - Human cost: degraded access to cloud services, real-time applications, and digital economy tools for populations with the least connectivity alternatives This platform answers: *where should new peering agreements be established, and what is the predicted regional latency dividend if they were?* --- ## Architecture Overview ``` ┌─────────────────────────────────────────────────────────┐ │ DASHBOARD (React + Vite) │ │ Topology Map │ Trombone Panel │ Fragility Rank │ Sim │ └──────────────┬──────────────────────────┬───────────────┘ │ REST (Tinybird pipes) │ WebSocket (optional, │ │ live BGP bridge) ┌──────────────▼──────────────┐ ┌────────▼──────────────┐ │ TINYBIRD │ …