Innovation

Building the Future of Humanitarian Intelligence

Systems and platforms I am actively designing to advance how humanitarian organisations collect, analyse, and act on data. Each project builds on a decade of operational experience to solve problems I’ve encountered firsthand.

AISA — Agentic Intersectoral Situational Analysis

Architecture & Prototyping

AI-powered humanitarian intelligence that operationalises intersectoral analysis across 14 UN agencies and 22 sectors.

The Problem

The humanitarian system processes over 100,000 field reports annually, yet only 5–10% receive structured analysis due to manual bottlenecks. Each agency operates siloed analytical workflows with no platform performing intersectoral analysis at the speed and scale required.

The Vision

An AI-powered platform using agentic workflows (15+ specialised agents) combined with ML, NLP, and large language models to automate the collection, harmonisation, analysis, and communication of humanitarian data — transforming months of manual analysis into hours.

Core Innovations

Multi-Agent AI Architecture — 15+ specialised agents orchestrated by LangGraph for document ingestion, classification, analysis, and report generation

Dual-Filter UX — Simultaneous agency-lens (14 UN agencies) and cluster-lens (22 sectors) enabling 308 view combinations

Analytical Confidence Engine — Cross-cutting quality scoring with confidence-weighted consensus metrics and information gap detection

Automated Document Generation — SitReps, country profiles, and briefing notes auto-generated in <15 minutes

Technology Stack

LangGraph + Claude APIPython FastAPIReact + TypeScriptPostgreSQL + Qdrant + Neo4jspaCy + Hugging FaceMapbox GL JS

Target Metrics

  • ▸90% reduction in qualitative analysis time
  • ▸>85% thematic classification accuracy
  • ▸5-page SitRep draft in <15 minutes
  • ▸308 agency×sector view combinations

Climate Anticipation Centre

Concept & Architecture

GeoAI-powered multi-hazard monitoring and anticipatory analytics — from passive spatial data to active climate intelligence.

The Problem

Climate disasters caused $4.3 trillion in losses (1970–2021), yet only 55% of WMO member states have functioning Multi-Hazard Early Warning Systems. Existing platforms are passive data repositories locked behind a technical gate.

The Vision

A next-generation GeoAI platform fusing autonomous GIS agents, AI-powered weather models, and multi-hazard analytics to autonomously monitor hazards, generate risk analytics, and communicate warnings in natural language.

Core Innovations

Autonomous GIS Agents — Specialised agents (Hazard Monitoring, Exposure Analysis, Climate Forecast, Risk Communication) using LLMs as reasoning cores

GeoAI Multi-Hazard Engine — AI weather foundation models (GraphCast, FourCastNet) for rapid 10-day forecasts in seconds

Anticipatory Action Packs — Auto-generated, hazard-specific response packages with action checklists, communication templates, and resource calculators

Natural Language Risk Querying — "Show me all districts at high flood risk in the next 72 hours"

Technology Stack

LangGraph + Claude APIGeoServer + PostGISGraphCast + FourCastNetReact/Next.js + Mapbox GL JSTensorFlow + PyTorchApache Kafka

Target Metrics

  • ▸8+ hazard types monitored
  • ▸≥24-hour early warning lead time
  • ▸AA-Packs generated in <15 minutes
  • ▸>80% adoption by non-GIS staff

ReportCentre

Concept & Design

Next-generation humanitarian reporting platform — evolving from data collection to evidence-based decision intelligence.

The Problem

Despite the success of platforms like ReportHub, significant gaps remain in how humanitarian data is collected, analysed, and communicated. Current systems operate in silos, lack advanced analytical capabilities, and struggle to translate raw 5W data into actionable strategic intelligence.

The Vision

An evolution of humanitarian reporting platforms that enhances how data is collected, analysed, and communicated — integrating modern data engineering, AI-powered analytics, and intuitive communication layers.

Core Innovations

Enhanced Data Collection — No-code relational database builder, offline-first data entry, multi-modal data ingestion

Advanced Analytics Engine — AI-powered pattern detection, predictive modelling, automated gap-overlap analysis

Intelligent Communication Layer — Automated insight generation, drag-and-drop report designer, role-based dashboards

Interoperability — Power BI, Python, R, ArcGIS integration pipelines with IATI and HDX compatibility

Technology Stack

React / Next.jsNode.js / FastAPIPostgreSQL + PostGIS + TimescaleDBPython (pandas, scikit-learn)Power BI EmbeddedDocker + Cloud-native

Target Metrics

  • ▸Built on ReportHub lessons (120+ partners)
  • ▸Real-time data updates
  • ▸AI-powered pattern detection
  • ▸Cross-platform interoperability

ReliefCash

Concept & Design

Intelligent cash transfer coordination — joint targeting, automated deduplication, and collective impact analytics.

The Problem

Multiple agencies deliver cash assistance to overlapping populations without real-time visibility into each other's activities, leading to duplication, coverage gaps, inconsistent transfer values, and inability to demonstrate collective impact.

The Vision

An intelligent cash transfer coordination system enabling humanitarian partners to work together through a joint targeting platform with automated deduplication, shared beneficiary registries, and evidence-based transfer value harmonisation.

Core Innovations

Joint Targeting System — Shared beneficiary registry with privacy-preserving deduplication and vulnerability-based prioritisation

Intelligent Cash Coordination — Real-time multi-agency dashboard, automated MEB calculation, FSP capacity mapping

Accountability & Fraud Prevention — Biometric verification, automated anomaly detection, end-to-end audit trail

Impact Analytics — Collective outcome measurement, PDM aggregation, cost-efficiency analysis, donor reporting automation

Technology Stack

PostgreSQL + privacy-preserving hashingPython (scikit-learn, pandas)React + MapboxPower BI EmbeddedReact Native (mobile)FastAPI + OAuth2

Target Metrics

  • ▸Built on CBIIMS foundation
  • ▸Privacy-preserving deduplication
  • ▸Real-time coordination dashboard
  • ▸Automated donor reporting

PDM Meta-Analysis Framework

Completed & Adopted

The first inter-agency PDM meta-analysis methodology — a 9-pillar framework unifying fragmented cash impact data.

The Problem

Cash Working Groups worldwide lack a standardised, reproducible method to measure collective impact across partners. Each partner uses different instruments, sampling approaches, and reporting formats.

The Vision

A replicable 9-pillar analytical methodology that transforms fragmented partner PDM data into unified inter-agency evidence, enabling cross-partner comparison, trend analysis, and aggregate impact measurement.

Core Innovations

Sample Profile & Demographics — Representation and geographic coverage analysis

Transfer Adequacy — Purchasing power against MEB benchmarks (39.1% fully met)

Experienced Outcomes — Satisfaction rates (93.8%), food security, coping strategies

Exclusion Signals — Protection risks (10.4% extortion rate), access barriers, coverage gaps

Technology Stack

Python (pandas, numpy, scipy)KoboCollect (152-question tool)Power BI (50+ DAX measures)statsmodelsGitHub (open-source pipeline)

Target Metrics

  • ▸1,559 households analysed
  • ▸5 organisations unified
  • ▸Adopted by Ethiopia CWG
  • ▸Open-source replicable pipeline

Interested in collaborating on any of these initiatives?

Let’s Talk