AI.info
ML data engineering
The evidence platform underneath every model: prediction time, ownership, contracts, and the leaks that look like accuracy.
- ML Data Engineering as an Evidence System
- Prediction Time, Unit of Analysis, and Outcome Horizon
- Source Discovery, Ownership, and Data Authority
- Semantic Contracts and Business Invariants
- Learning Event Instrumentation and Taxonomy
- Clocks: Event, Ingestion, Availability, and Label Time
- Batch, Change Data Capture, and Stream Ingestion
- Windows, Watermarks, and Late Data
- Storage Layers, File Formats, and Table Layout
- Table Snapshots, Compaction, and Retention
- Schema Evolution, Migrations, and Historical Backfills
- Identifiers, Keys, and Entity Resolution
- Joins, Cardinality, Deduplication, and Reconciliation
- Training Dataset Construction and Release
- Point-in-Time Correctness and Historical Retrieval
- Labels, Ground Truth, and Annotation Pipelines
- Label Delay, Censoring, and Proxy Targets
- Data Profiling and Statistical Baselines
- Data Quality Rules, Constraints, and Release Gates
- Missingness, Outliers, Corruption, and Repair
- Feature Transformations and Stateful Pipelines
- Data Engineering for Text, Images, Audio, and Sensors
- Categorical, Sparse, and High-Cardinality Data
- Split Design, Sampling, and Leakage Prevention
- Imbalanced Data, Rare Events, and Coverage
- Time-Series and Sequential Data Engineering
- Graph Data Engineering and Dynamic Relations
- Data Augmentation, Synthetic Data, and Weak Supervision
- Feature Stores and Historical Retrieval
- Online Serving Data, Caching, and Fallbacks
- Dataset Versioning, Lineage, Provenance, and Reproducibility
- Orchestration, Idempotency, Backfills, and Recovery
- Data Validation and Release Gates
- Data Observability, SLIs, and Incident Response
- Drift, Freshness, Refresh, and Dataset Retirement
- Privacy, Consent, Retention, and Deletion
- Data Security, Poisoning, and Supply-Chain Risk
- Dataset Documentation, Governance, and Use Review
- Cost, Performance, and Data Platform Design
- Capstone: Design and Defend an ML Evidence Platform