Feat Mesh 1.0 API Documentation

Overview

Feat Mesh 1.0 exposes the FEATSRV research stack through a browser console and authenticated REST surface. Clients fetch online features by entity id, build leakage-free training frames at prediction timestamps, audit temporal correctness, and inspect ranked feature importance across actuarial models.

Online Redis — policy and claim feature lookups under 10 ms
PIT training — as-of joins that never see future feature values
Leakage audit — flag feature_timestamps after event_timestamp

Authentication

Requests accept an x-api-key header. The public demo console uses a research prototype key suitable for evaluation traffic. Production deployments should rotate keys via environment configuration (FEATMESH_DEMO_API_KEY / upstream backend policy).

x-api-key: <your-api-key>
Content-Type: application/json
Accept: application/json

Same-origin proxy

Browser clients should call the Æ Hive proxy at /api/featmesh/*. The route forwards to the feature-serving backend and injects the demo key when none is supplied. When the backend is unreachable, the console falls back to static Nimbus demo JSON under /investor-relations/research/point-in-time/data/.

GET  /api/featmesh/health
GET  /api/featmesh/online/policy/{id}
GET  /api/featmesh/online/claim/{id}
POST /api/featmesh/offline
POST /api/featmesh/pit-training
POST /api/featmesh/leakage-audit
GET  /api/featmesh/feature-importance

Endpoints

Method	Path	Purpose
`GET`	`/health`	Service health and store status
`GET`	`/online/policy/{id}`	Real-time policy feature lookup
`GET`	`/online/claim/{id}`	Real-time claim feature lookup
`POST`	`/offline`	Batch offline feature materialization
`POST`	`/pit-training`	Leakage-free point-in-time training set
`POST`	`/leakage-audit`	Detect future feature timestamps
`GET`	`/feature-importance`	Ranked importance across model families

GET /online/policy|claim/{id}

Returns the latest online feature vector for a policy or claim entity from the Redis-backed store. Used by underwriting and fraud scoring paths that need sub-10 ms feature materialization.

GET /api/featmesh/online/policy/POL-10042
Accept: application/json
x-api-key: <your-api-key>

POST /pit-training

Builds a leakage-free training frame by joining entity rows at each event_timestamp against SCD Type 2 feature history. Only feature values valid as-of the prediction time are returned.

{
  "entities": [
    { "claim_id": "CLM-001", "event_timestamp": "2024-06-15T12:00:00Z" }
  ],
  "features": ["claim_severity_score", "prior_claims_24m"]
}

POST /leakage-audit

Scans joined feature rows for feature_timestamp values that fall after the corresponding event_timestamp, returning severity-ranked findings for actuarial portfolio review.

Point-in-time feature serving fundamentals