Build production-ready anomaly detection systems using ResNet embeddings for observability data.
Topics Covered¶
ResNet Architecture: Understanding residual networks and adapting them for tabular data
Feature Engineering: Transforming OCSF observability data into model-ready features
Self-Supervised Learning: Training on unlabeled data using contrastive learning
Embedding Quality: Evaluating embeddings with quantitative and qualitative methods
Anomaly Detection: Applying distance-based, density-based, and ensemble methods
Production Deployment: FastAPI services, vector databases, and real-time inference
MLOps: Monitoring, drift detection, and automated retraining
Multi-Source Correlation: Root cause analysis across logs, metrics, and traces
What You’ll Build¶
A complete production system including:
Custom TabularResNet model trained with self-supervised learning
Embedding service (FastAPI) for real-time inference
Vector database for fast k-NN similarity search
Anomaly detection through vector operations
Monitoring and automated retraining pipeline
Access the Tutorial Series¶
Observability Anomaly Detection →
The complete tutorial series is available as an interactive MyST book with hands-on notebooks, sample data, and production-ready code examples.
Target Audience¶
ML engineers building anomaly detection systems
Security engineers working with observability data
Data scientists interested in self-supervised learning
Anyone applying deep learning to tabular/observability data
Prerequisites¶
Basic Python and PyTorch
Understanding of neural networks (or see our Neural Networks From Scratch series)
Key Features¶
✅ Production-ready - All examples are deployable ✅ No labels required - Self-supervised learning on unlabeled data ✅ Hands-on notebooks - MyST markdown format with executable code ✅ Sample data included - Pre-generated OCSF events ✅ Complete MLOps - Deployment, monitoring, retraining ✅ Broadly applicable - Works with any structured observability data