Home

Technical Documentation for Humanitarian AI Systems in Uganda

---

Overview

This framework provides a comprehensive approach to optimizing health supply chains in low-resource humanitarian settings. Developed by the International Foundation for Recovery and Development (IFRAD) under the Elrha Humanitarian Innovation Fund, it addresses the unique challenges of Uganda's health system where 89% of facilities experience unreliable connectivity and infrastructure constraints directly impact medicine availability.

Core Design Principles

Principle	Implementation
Offline-First	All critical functions work without internet connectivity
Human-in-the-Loop	AI provides recommendations; health workers make final decisions
Infrastructure-Aware	Storage capacity treated as a hard constraint, not a variable
Paper-Digital Harmony	System augments, not replaces, existing paper workflows

Three-Tier Forecasting Architecture

┌─────────────────────────────────────────────────────────────┐
│                    CENTRAL DATA HUB                         │
│         National reporting & system integration             │
└─────────────────────────────────────────────────────────────┘
                              ▲
                              │ Periodic ETL
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                   DISTRICT SERVICES                         │
│    Tier 2: Hierarchical Statistical Forecasting (HES)       │
│    Tier 3: Machine Learning Models (Random Forest/XGBoost)  │
└─────────────────────────────────────────────────────────────┘
                              ▲
                              │ Opportunistic Sync
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                   FACILITY LEVEL                            │
│         Tier 1: Rule-Based Forecasting (Offline)            │
│              SQLite + AES-256 Encryption                    │
└─────────────────────────────────────────────────────────────┘

Key Baseline Findings

The framework directly addresses findings from the 2025 Baseline Assessment:

• Storage-Stockout Correlation: r = -0.695 (strong negative)

• Connectivity Failure Rate: 89% of facilities

• Stockout Prevalence: 100% of pilot facilities experienced stockouts

• Maximum Stockout Duration: 120 days at refugee-serving facilities

---

Documentation Structure

Technical Specifications

• System Architecture - Distributed system design for offline-first operations

• Predictive Demand Algorithm - Three-tier forecasting methodology

• Implementation Manual - Android-specific development guide

• Offline Workflow - Interface design and data flow

Integration & Standards

• Integration Standards - Maturity model and system interoperability

Operations

• Decision-Making Library - Override protocols and use cases

• Fraud Prevention Handbook - Detection mechanisms and audit trails

Ethics & Compliance

• Ethical Guidelines - Humanitarian AI principles

Assessment

• Value for Money - Quantified impact assessment

---

Project Information

Field	Details
Lead Organization	International Foundation for Recovery and Development (IFRAD)
Funding	Elrha Humanitarian Innovation Fund (UK FCDO)
Pilot Region	Karamoja and Southwestern Uganda
Pilot Facilities	10 (2 Regional Referral Hospitals, 8 Health Centers)
Phase	Phase 1 Pilot (July - December 2025)

---

Version 2.0 - Post-Validation Revision | December 2025