axmedmaxamuud/Somalia-Conflict-Monitor

AI-powered conflict, humanitarian and climate overlay analysis for Somalia # Somalia Conflict Monitor An automated conflict monitoring tool that integrates five humanitarian data sources and uses AI to generate monthly analytical briefs in the style of intelligence products. The pipeline runs weekly, pulls current data, produces a structured narrative analysis, and publishes outputs without manual intervention. --- ## Live Dashboard somalia-conflict-monitor.streamlit.app --- ## Data Sources - **ACLED** — Armed Conflict Location and Event Data; disaggregated event-level conflict records updated weekly - **IPC** — Integrated Phase Classification; population-level food security phase assessments by region - **CHIRPS** — Climate Hazards Group InfraRed Precipitation with Station data; dekadal rainfall anomaly against the 1989–2018 baseline - **UNFPA** — 2021 population projections aggregated to admin1, used for per-capita conflict rate calculations - **IOM DTM / UNHCR-OCHA** — Harmonised IDP figures from the Humanitarian Data Exchange, supplemented by IOM DTM API data --- ## Architecture The pipeline pulls data from five sources via API, compiles event-level and aggregate datasets, and passes them to a large language model with a structured analytical prompt. Outputs are produced in three formats: a Word document, a standalone HTML file, and a Streamlit dashboard. A GitHub Actions workflow runs every Monday at 06:00 UTC, commits updated data files, and triggers a fresh dashboard deployment. --- ## Methodology The narrative analysis is generated by Claude (Anthropic). The model receives event-level ACLED data for the reporting month alongside IPC, CHIRPS, population and displacement datasets. It is instructed to derive conclusions from the data only, distinguish factual statements from analytical commentary using explicit comment blocks, present competing assumptions for significant claims, use calibrated probability language, and cite ACLED event IDs as footnote references. A post-processing filter removes superlative constructions …