CreditSetu

General-purpose, explainable credit intelligence engine for thin-file and alternative-data lending. Score customers on behavioral transaction data — including customers with no bureau score at all — using three composable engines (Intent, Capacity, Guardrail), SHAP-backed explanations, and regulator-style adverse action reason codes.

This is the library package. For the full project — a reference FastAPI service, a React dashboard, synthetic demo data, and deployment docs — see the GitHub repository.

Install

pip install creditsetu

That pulls in only pandas, numpy, scikit-learn, lightgbm, pydantic, and pyyaml — enough to fit models and score customers. SHAP explainability and ruptures-based life-event detection are opt-in and degrade gracefully if skipped:

pip install "creditsetu[explain]"   # + SHAP feature attributions / adverse action reasons
pip install "creditsetu[intent]"    # + ruptures change-point life-event detection
pip install "creditsetu[full]"      # both

Usage

from app.pipeline import CreditIntelligencePipeline
from app.scoring_profile import ScoringProfile

# Currency, guardrail thresholds, product catalog, weights — all policy,
# no code changes needed to adapt this to a different institution/market.
profile = ScoringProfile(name="acme", org_name="Acme Lending", currency_symbol="$")
pipeline = CreditIntelligencePipeline(profile=profile)

# Fit on your own historical data — customers_df needs a
# `true_repayment_capacity` column, ideally a real `is_stressed` column too.
pipeline.fit(customers_df, transactions_df)

# Score one customer...
result = pipeline.score_customer(customer, customer_transactions_df)

# ...or a whole batch, vectorized end-to-end.
scores_df = pipeline.score_batch(customers_df, transactions_df)

Data contract — customer is a dict needing at least customer_id and bureau_score (None is a first-class value, not an edge case — that's the point). transactions is a DataFrame with date, amount, type (credit/debit), category, counterparty, is_bounce.

Full details live in the package's pipeline.py module docstring, and the complete architecture/evaluation writeup is in the GitHub README.

License

MIT