SentryML
ML observability & MLOps — model monitoring, drift detection, debugging in production.
The ML Monitoring Metrics Taxonomy: Drift, Data Quality, and Model Decay
A reference taxonomy of the signals that actually tell you a production ML system is failing — input drift, prediction drift, concept drift, data quality, and performance decay — and which open-source tool computes each one.
A Field Guide to the OpenTelemetry GenAI Semantic Conventions
What the OpenTelemetry GenAI semantic conventions actually standardize — spans, events, and metrics for LLM and agent telemetry — what they don't yet cover, and how to instrument an LLM app against a moving spec without painting yourself into a corner.
Model Monitoring in Production: A Four-Layer Framework
Model monitoring covers more than drift detection. Here's the four-layer framework — software health, data quality, model quality, business KPIs — wired up with Evidently, PSI thresholds, and real alert logic.
Earlier entries
-
Model Monitoring for LLM Inference: Metrics Your APM Can't See
monitoring -
SmithDB and Five Other Things LangChain Shipped at Interrupt 2026
tooling -
LLM Benchmarks in 2026: Which Still Discriminate, and How to Run
mlops -
Watermarking Should Be Treated as a Monitoring Primitive
monitoring -
LLM Fine Tuning: Methods, Training Data, and Evaluation
mlops -
LLM Testing: A Guide to Evals, Metrics, and Production Monitoring
monitoring -
ML Testing: A Checklist from Pre-Train Checks to Production Drift
mlops -
Choosing MLOps Tools: A Decision Framework for Production Teams
mlops -
When Embedding-Based Defenses Fail in Multi-Agent LLMs
deep-dive -
LLM Benchmarks Explained: What the Numbers Mean and Miss
mlops
Trusted by researchers across the AI security community
SentryML is part of a 26-site editorial network covering adversarial ML, AI governance, defensive tooling, and ops engineering — all open access.
SentryML — in your inbox
ML observability & MLOps — model monitoring, drift detection, debugging in production. — delivered when there's something worth your inbox.
No spam. Unsubscribe anytime.