Real-Time Visual Anomaly Detection for Steel Surface Inspection
An edge-to-MES architecture that catches surface defects at line speed using PatchCore anomaly scoring and a TensorRT inference path on the plant floor.
Read article
Technical deep-dives from the Datavriksh engineering team — how we design, build, and operate production AI systems for heavy industry. Real architectures, real trade-offs, real code.
An edge-to-MES architecture that catches surface defects at line speed using PatchCore anomaly scoring and a TensorRT inference path on the plant floor.
Read articleChunking, hybrid BM25 + dense retrieval, cross-encoder reranking, grounded prompting with citations, and an evaluation harness that gates every release.
Read articleHow to train, serve, monitor, and retrain hundreds of thousands of models with a feature store, a model registry, and drift-triggered pipelines.
Read articleA gateway architecture with prompt versioning, CI evaluation gates, input/output guardrails, semantic caching, and token-level cost observability.
Read articleA streaming pipeline that fuses IT and OT telemetry, scores multivariate anomalies with an LSTM autoencoder, and correlates alerts over a topology graph.
Read articlePoint-in-time-correct features, external regressors (weather, holidays), gradient-boosted quantile models, and hierarchical reconciliation at scale.
Read articleOur engineers are happy to talk architecture for your environment.
Talk to our team