Built-in reference models
Reference models that ship with the API, so you can try every endpoint without deploying anything.
Five reference models ship with the API so you can exercise every endpoint without deploying anything. Register one with inference: {"type": "builtin", "name": …}; Kaimei fills in the features, outputs and task, and kaimei-api seed-demo gives each one reference data, traffic and insights.
| name | Task | Inputs | What it is |
|---|---|---|---|
credit-risk | binary_classification | 9: income, credit_score, utilization, debt_to_income, loan_amount, employment_years, num_delinquencies, age, home_ownership | Consumer-credit decline model with a utilisation × debt-to-income interaction and an accelerating sub-prime penalty. Outputs approve / decline. |
energy-demand | regression | 7: temperature, hour, day_type, humidity, wind_speed, solar_irradiance, industrial_index | Hourly electricity demand in MW with heating/cooling asymmetry and daily cycles. |
churn-risk | binary_classification | 6: tenure_months, monthly_spend, support_tickets, days_since_login, sessions_30d, failed_payments | A small subscription-churn network. Outputs stay / churn. |
equity-alpha | regression | 8: momentum_12_1, earnings_yield, roe, volatility, log_market_cap, earnings_revision, short_interest, turnover | A three-layer network scoring expected one-month residual return, in % per month. |
cats-dogs | binary_classification | 1 image: 16 × 16 × 3 | A convolutional network that classifies a 16×16 colour drawing as cat or dog. Supports Grad-CAM. |
GET /v1 lists them with their feature counts. They are models like any other once registered: explain them, analyse them, log predictions against them, delete them.