See Problems Before They Become Incidents
Real-time anomaly detection across services, pipelines and integrations — every decision SHAP-explained, every model rollback-tested, every alert carrying its evidence.
Problems are what you can see in time. Sentinel watches every service, pipeline and integration that touches a FiduciaWay platform, scores behavior in real time, and attaches a SHAP attribution to every prediction — so when an operator asks why a job was flagged, the answer is a query we already have, against data we already trust. Inference runs under 200ms p99 on the alert path, because an anomaly signal that times out is an anomaly signal that missed. The holdout set is immutable once a model is in production, bias is tested on the workloads the model scores, and drift beyond a threshold reverts the model to its previous version automatically.
We refuse the superstition that a deep neural network is always the answer. On an alert path, an explainable gradient-boosted model that an operator can interrogate beats a marginally-more-accurate black box we cannot defend. Features are derived from the audit log, versioned and reproducible; PII stays in the ledger, encrypted and access-listed, and the feature store holds only derived, minimised features. An operator's "why" is a query we already have; the dashboard is a convenience, the audit log is the contract. Operations owns the questions; we own the answers.
Key Capabilities
Stream Anomaly Detection
Behavioral baselines across services, queues and data pipelines; inference p99 under 200ms on the alert path.
Explainable Decisions
SHAP attributions attached to every prediction; an operator can interrogate any flag without a data scientist in the room.
SLO & Error-Budget Tracking
Every service and pipeline carries an SLO; error-budget burn alerts before customers feel the burn.
Champion/Challenger Evaluation
New models compete against production on immutable holdouts; drift beyond threshold reverts automatically.
Built With
p99 decision latency
Monitoring coverage
Rollback on model drift
How It’s Built & Delivered
Architecture
Streaming feature pipeline over the audit log; every model decision recorded as an event you can replay.
Verification
Bias, drift and rollback behavior covered by release-gated tests; holdout sets immutable once in production.
Delivery
Every release rides the gated pipeline: review, scans, signed provenance, canary, automatic rollback.
Success Stories
Nile Logistics Group
Declaration backlog detected 47 minutes before a carrier SLA breach; 11 incidents avoided in year one
Sahara Retail Group
Inventory-sync anomalies caught pre-launch; false positives down 71% after champion/challenger rollout
Sentinel questions
Why was my job flagged?
Every prediction carries SHAP attributions; the top drivers, their direction, and the threshold are one query away — no data scientist required.
How fast is detection?
p99 under 200ms on the alert path; a signal that times out is a signal that missed. We treat inference latency like product latency.
Can operations write its own rules?
Yes — a rule studio in plain language, compiled deterministically; rules and models are versioned, tested, and rollback-ready.
What happens when a model drifts?
Drift beyond a threshold reverts the model to its previous version automatically and pages the owner with the evidence attached.
Does it watch data pipelines too?
Yes — schemas, volumes, latencies and reconciliation outcomes; the same explainability contract as services.