Research Analyst & Data Scientist · IFPRI / CGIAR · Dakar, Senegal
Reproducible pipelines and tools that turn household surveys, disease surveillance, satellite imagery and conflict data into evidence for food security and humanitarian decisions in the Sahel and Sub-Saharan Africa.
Config-driven, tested workflow that harmonises Chad's national food security survey rounds into comparable thematic datasets, recomputes indicators and validates them against the original analysts' scores.
R pipeline on weekly IDSR data: automated epidemiological bulletin, French-language dashboard (malaria, Ebola, data quality) and a controlled interrupted time series across health zones. Runs end to end on synthetic demo data.
run_all.R --demoTen years of weekly notifiable disease reports cleaned, validated against source totals and served in a Shiny dashboard with meningitis alert thresholds, reporting completeness and a DHIS2 comparison, plus count and endemic-epidemic models.
R package and Shiny app that makes species distribution modelling accessible to non-programmers, with spatially blocked cross-validation, predictor screening and export of the generated R code. Example data: agroforestry trees in Niakhar, Senegal.
Google Earth Engine workflow mapping arable land by combining cropland extent and cleared forest, then masking steep slopes, permanent water and impervious surfaces. One script per country.
Eleven practical sessions taught at ENSAE Dakar: raster and vector data, zonal statistics, risk mapping with WorldPop and ACLED, spatial autocorrelation, species distribution models and Sentinel-2 indices.
Ask in English or French how Chad's 14 ENSA rounds were harmonised and what they show. Read-only tools trace any variable across rounds, explain recodes and quality checks, and compare indicators by round, region and sex of the household head. No household record reaches the model.
Ask in English or French how the 10 survey rounds of WFP's Integrated Resilience Programme were merged, which checks fail, which data issues remain open, and what the indicators show by round and region, with WFP's reconciled costs. Charts that join lean-season and post-harvest rounds are refused.
Plain-language questions on Niger's weekly notifiable disease surveillance, answered with tables, charts, maps and CSV. The tools enforce epidemiological rules: like-for-like year comparisons, missing reports never read as zero, case-definition changes always flagged.
Question-and-answer agent on Ituri's IDSR data, packaged as a Claude Code skill. Indicators come from the R pipeline, never from the model, with a parity test against the pipeline. Real data stay locked until a written data-sharing agreement exists. Work in progress.
AI startup building decision tools for food security, poverty targeting and agricultural transformation in Africa, combining satellite data, household surveys and machine learning.
Household dataset for IFPRI's cost-benefit analysis of WFP's Integrated Resilience Programme in Chad, 2018 to 2023: 26 modules merged, WFP costs reconciled by activity, a Shiny dashboard of checks and results, and a package for public release on IFPRI Dataverse.
Reproducible pipeline for the MSF Ebola treatment centre in Mongbwalu, Ituri: weekly sitrep, line-list analysis report, daily data quality checklist and DHIS2 analyses (epidemic curves, transmission chains, Rt). Patient data stay on the analyst's machine; CI runs on synthetic data. Code © MSF OCG.
Exposure measures for survey clusters (10, 25 and 50 km) and the 519 Admin 3 units of the DRC: ACLED conflict events, WorldPop population and age-sex structure, and ERA5-Land temperature anomalies, validated against UCDP GED. Outputs in CSV and Stata.
Data acquisition for the Mastercard Foundation agrifood scenarios study with IFPRI: a verified source matrix and access register for labour force and household surveys, with scripts that validate the register, build the Excel deliverables and keep microdata out of the repository.
How households respond to conflict, climate and price shocks, using eleven waves of cross-sectional survey data linked to an extensive set of external shock indicators.
Companion analysis for Mali, using association rule mining to identify which coping strategies households combine under different shocks and stressors.
ACLED's conflict diffusion indicator implemented in Earth Engine: share of populated 10 km cells with high violence, from WorldPop and ACLED, with sensitivity analysis on thresholds. Prepared for WFP.
Poverty headcount below the survey's representative levels, comparing direct, Fay-Herriot and machine learning estimates with conflict and climate covariates. Series for six West African countries.
Provenance-tracked pipeline assembling sources on geography, soil, food, health and policy into one database keyed on a county crosswalk, for soil-to-nutrition pathway analysis.
Guide to replicating the Fields of The World 10 m field boundary map for Burkina Faso: Sentinel-2 mosaics, PRUE U-Net inference, polygon extraction and confidence filtering.
Field data collection for low-connectivity areas, built by extending ODK: ODK Central hosted in Africa on AWS, a branded ODK Collect and an encrypted SMS fallback. The SMS codec and bridge service are built and tested; infrastructure is written but not yet deployed.
Supervision tools for the ANSD beehive productivity survey collected with Survey Solutions: fieldwork progress by day, region and enumerator, consistency checks, and a ready-to-use list of interviews to reject. R Shiny and Python versions give identical results.
Training material prepared for the International Fertilizer Development Center on data handling and analysis in Excel.
Introductory SPSS course material in French, prepared at ENSAE Dakar for students starting with statistical software.