Serve actuarial features with point-in-time correctness
FeatServe 1.0 exposes FEATSRV online Redis lookups, offline PIT training sets, leakage audits, and feature importance — a live console on Kryptur infrastructure for underwriting, pricing, and fraud models.
Recommended for you
Top research capabilities
Kryptur advances smarter actuarial MLOps through point-in-time joins, Redis online serving, leakage audits, and a seven-service CLI/dashboard stack.
FeatServe 1.0™
Live feature console — overview, lineage, risk, importance, training, and online lookups.
Online Redis
Real-time policy and claim feature lookups averaging under 10 ms on the deployed VPS.
PIT training
File-upload and batch offline joins return leakage-free DataFrames for model training.
Leakage audit
Detect future feature_timestamps relative to event_timestamp across actuarial portfolios.
POST /api/featserve/*
Same-origin proxy to FEATSRV FastAPI — health, online, offline, pit-training, importance.
Nimbus packs
Seven CLI service packs with intense demo CSVs for every actuarial workflow.
Smarter impact powered by Kryptur
Redis-backed policy and claim lookups feed underwriting and fraud models in real time.
Offline point-in-time joins on 1,000 claims complete in under 2 seconds locally.
PySpark batch path completes the same 1,000-claim join in under 5 seconds.
Seven CLI services cover training, online, offline, leakage, importance, and health.
Same-origin /api/featserve forwards authenticated requests to the FEATSRV backend.
Unlock expert insights on leakage-free feature serving
Optimise actuarial ML outcomes from point-in-time stores and online Redis investments.
API FeatServe 1.0 · GET /online/* · POST /pit-training
Same-origin proxy at /api/featserve forwards JSON to the FEATSRV FastAPI backend. Endpoints include /health, /online/policy/{id}, /online/claim/{id}, /offline, /pit-training, /leakage-audit, and /feature-importance. Auth via x-api-key. Console falls back to Nimbus demo JSON when live backends are unreachable.
Feature serving, without the leakage.
A visual companion to the FEATSRV CLI. Every panel below corresponds to one of the CLI's seven services — point‑in‑time training sets, online policy/claim lookups, batch offline features, leakage audits, feature importance, and system health — rendered from the same JSON responses the terminal saves to Response‑FEATSRV/.
Online feature request volume — before / after real‑time rollout
Monthly request count against the Redis‑backed online store. The dashed line marks the cut‑over from batch‑only serving (Jan 2024) to real‑time point‑in‑time serving.
Recent CLI activity
| Timestamp | Service | Entity | Status | Latency |
|---|---|---|---|---|
| Loading real activity… | ||||
Service call mix (24h)
Feature lineage — source to consumer
Every value served online or offline traces back through the point‑in‑time store. Flow width is proportional to feature‑value volume over the trailing 30 days.
Leakage exposure by feature domain
Percent of served feature‑values at risk of temporal leakage.
Leakage risk by portfolio
Projected rows requiring regeneration
Leakage trend
Entity risk‑score distribution
Open leakage audit findings
| Entity id | Feature | Portfolio | Δ hours | Severity |
|---|---|---|---|---|
| POL-3217 | claim_severity_score | Workers Comp | -3.3 | medium |
| POL-5959 | exposure_months | Auto Comprehensive | -6.8 | medium |
| CLM-86657 | claim_severity_score | Auto Liability | -11.9 | medium |
| CLM-85589 | prior_claims_24m | Homeowners | -8.5 | medium |
| CLM-79370 | region_risk_index | Auto Liability | -1.8 | high |
| CLM-74746 | exposure_months | Auto Comprehensive | -2.4 | medium |
| CLM-62267 | telematics_score | Workers Comp | -7.8 | medium |
| POL-9634 | claim_severity_score | Workers Comp | 3.4 | high |
| POL-5211 | prior_claims_24m | Auto Comprehensive | 3.9 | high |
| CLM-93983 | region_risk_index | Workers Comp | -7.3 | medium |
| CLM-90742 | bureau_score_v3 | Auto Liability | -10.6 | medium |
| CLM-80942 | prior_claims_24m | Umbrella | -0.3 | high |
| POL-8171 | claim_severity_score | Auto Comprehensive | -3.8 | medium |
| CLM-23712 | telematics_score | Umbrella | 0.0 | high |
| POL-2352 | claim_severity_score | Umbrella | -8.9 | medium |
| CLM-29388 | bureau_score_v3 | Workers Comp | -5.5 | medium |
| POL-4688 | exposure_months | Auto Liability | -45.1 | low |
| CLM-27431 | claim_severity_score | Umbrella | -37.1 | low |
| CLM-66387 | exposure_months | Auto Comprehensive | -31.0 | low |
| CLM-58003 | bureau_score_v3 | Auto Liability | -27.0 | low |
Feature importance summary
Ranked by mean absolute contribution across active underwriting and pricing models.
Top 14 features
Point‑in‑time training set preview
First rows of the most recent PIT training pull.
| claim_id | policy_id | event_timestamp | feature_timestamp | claim_severity_score | prior_claims_24m | label_charge_off |
|---|---|---|---|---|---|---|
| Loading PIT data… | ||||||
Batch offline features — last payload
Response body of the most recent /offline POST.
| entity_id | entity_type | telematics_score | credit_band | exposure_months | region |
|---|---|---|---|---|---|
| Loading batch data… | |||||
Online feature lookup
Simulates real‑time Redis‑backed lookups. Enter an ID or pick a recent entity.
Policy features
Claim features
Stay connected
What's New at Kryptur — FeatServe 1.0 API releases, point-in-time research, and actuarial MLOps news.