CreditPulse-Causal AI credit risk platform for East African digital lending
# CreditPulse
> **Causal AI credit risk assessment for East African digital lending markets.**
> Built for thin-file borrowers, M-Pesa ecosystems, and CBK regulatory compliance.
---
## What is CreditPulse?
CreditPulse is an open-source credit risk intelligence platform that goes beyond
correlation-based scoring. It asks not just *who* will default, but *why* —
using causal inference, Bayesian uncertainty quantification, and explainable AI.
### Key Results
| Model | Metric | Value |
|-------|--------|-------|
| XGBoost Classifier | AUC-ROC | 0.7415 |
| XGBoost Classifier | Gini Coefficient | 0.4830 |
| Random Survival Forest | C-index | 0.8130 |
| Difference-in-Differences | Causal coefficient | -0.1125 (p=0.002) |
| Regression Discontinuity | LATE at cut-off | 0.0058 (p=0.0028) |
| Fraud Detector | Applications flagged | 1.75% |
---
## Quickstart
```bash
# 1. Clone
git clone
github.com
cd creditpulse
# 2. Environment
conda create -n creditpulse python=3.11 -y
conda activate creditpulse
pip install -e .
pip install pandas numpy pyarrow xgboost==2.1.4 scikit-learn joblib shap
pip install scikit-survival dowhy econml statsmodels networkx matplotlib seaborn
pip install langchain langchain-community langchain-text-splitters
pip install langchain-huggingface chromadb bertopic sentence-transformers
pip install fastapi uvicorn streamlit pydantic httpx
# 3. Run the pipeline
python creditpulse/ingestion/clean.py
python creditpulse/causal/dag.py
python creditpulse/causal/diff_in_diff.py
python creditpulse/causal/regression_discontinuity.py
python creditpulse/nlp/rag_pipeline.py
python creditpulse/nlp/topic_model.py
python creditpulse/models/xgboost_model.py
python creditpulse/models/bayesian_model.py
python creditpulse/models/survival_model.py
python creditpulse/models/fraud_detector.py
python creditpulse/explainability/fairness_audit.py
python creditpulse/explainability/model_card.py
# 4. Launch
# Terminal 1
uvicorn creditpul …