FeatServe 1.0 · API Documentation
FeatServe 1.0 is the Kryptur console and same-origin REST proxy for FEATSRV — an open-source point-in-time feature store for insurance and actuarial ML. It guarantees temporal correctness so training and online inference never see future feature values.
Every feature value is joined as-of an event_timestamp. SCD Type 2 history (for example address or credit band) is resolved so only information available at prediction time enters the training row — eliminating look-ahead leakage.
Policy and claim lookups serve low-latency features from Redis (typically under 10 ms on the deployed VPS). Use GET /online/policy/{id} and GET /online/claim/{id} via the same-origin proxy.
POST /pit-training and POST /offline materialize leakage-free DataFrames for underwriting, pricing, and fraud models — backed by Feast, Dask, and PySpark engines in FEATSRV.
POST /leakage-audit flags rows where feature_timestamp is after event_timestamp, ranked by severity across actuarial portfolios.
Requests to /api/featserve/* forward to the FEATSRV FastAPI backend. Pass the API key with the x-api-key header. The console falls back to Nimbus demo JSON when live backends are unreachable.
curl -H "x-api-key: YOUR_KEY" \ https://www.kryptur.com/api/featserve/health
Proxy base: /api/featserve — all paths below are relative to the Kryptur origin.
| Method | Path | Summary |
|---|---|---|
| GET | /api/featserve/health | Service health and store status (Redis + parquet). |
| GET | /api/featserve/online/policy/{id} | Real-time policy feature lookup from the online Redis store. |
| GET | /api/featserve/online/claim/{id} | Real-time claim feature lookup for underwriting and fraud models. |
| POST | /api/featserve/pit-training | Build a leakage-free point-in-time training set from entity timestamps. |
| POST | /api/featserve/offline | Batch offline feature materialization for actuarial ML pipelines. |
| POST | /api/featserve/leakage-audit | Detect feature timestamps that leak information after the event timestamp. |
| GET | /api/featserve/feature-importance | Ranked feature importance across underwriting, pricing, and fraud models. |