This repository serves a demo for the conflict in Africa project with Yujing Huang, only a slice of data (Kenya) is displayed. Full version of code and data is not disclosed here.
# conflict_africa_demo — AI-Integrated Research Workflow Demo
A standalone demonstration of an AI-integrated empirical research workflow, built around a public ACLED conflict-event slice for Kenya, 2010-2020. The repo shows:
1. **Reproducible R pipelines** with explicit verification steps
2. **AI workflow infrastructure** — custom rules, slash commands, and persistent memory used to collaborate with Claude Code
3. **Engineering discipline** in data cleaning, aggregation, and validation
> No research findings are reproduced; the analysis stops at descriptive material.
## Quick tour
| What you're looking for | Where to find it |
|---|---|
| How I write reproducible R | `code/` — numbered pipeline `d_01`-`d_04` |
| How I use AI in my workflow | `docs/ai-workflow.md` |
| Custom Claude Code rules | `.claude/rules/` |
| Custom slash commands | `.claude/skills/` |
| Persistent memory pattern | `.claude/memory/MEMORY.md` |
| Pipeline overview | `docs/architecture.md` |
| End-to-end runbook | `CLAUDE.md` |
## What the pipeline does
```
d_01_acled_clean.R Clean and deduplicate the raw ACLED Kenya CSV
→ data/Conflict/ACLED/acled_kenya_clean.rds
d_02_acled_aggregate.R Build a year x admin1 x event_type panel of
event counts and total fatalities
→ data/Conflict/ACLED/acled_kenya_panel.rds
d_03_acled_check.R 11 verification checks (identity, key uniqueness,
count reconciliation, types, year coverage)
→ exit 0 = pass
d_04_descriptive_figure.R → output/figure_1_kenya_acled_events.png
```
## Run it
Requirements: R ≥ 4.2, packages: `here`, `readr`, `dplyr`, `tidyr`, `ggplot2`.
```bash
git clone
github.com
cd conflict_africa_demo
Rscript code/d_01_acled_clean.R
Rscript code/d_02_acled_aggregate.R
Rscript code/d_03_acled_check.R # → "11 PASS, 0 FAIL"
Rscript code/d_04_descriptive_figure.R # → output/figure_1_kenya_acled_events.png
```
All paths are relative to the repo root. The pipeline is fully sandboxe …