ai Directory Reference

Directories
	compliance

Files
	audit_chain.hpp
	column_batch.hpp
	Column-major batch of feature rows for zero-copy tensor wrapping and WAL serialization in ML inference pipelines.
	data_classification.hpp
	decision_log.hpp
	event_bus.hpp
	Multi-tier event bus for routing SharedColumnBatch events through three tiers.
	feature_reader.hpp
	Versioned ML feature store reader with point-in-time queries.
	feature_writer.hpp
	Versioned ML feature store writer for appending typed feature vectors to Parquet.
	human_oversight.hpp
	incident_response.hpp
	inference_log.hpp
	log_retention.hpp
	mpmc_ring.hpp
	Lock-free bounded MPMC ring buffer based on Dmitry Vyukov's algorithm.
	quantized_vector.hpp
	INT8/INT4 quantized vector storage for AI embeddings with SIMD acceleration.
	regulatory_monitor.hpp
	row_lineage.hpp
	Per-row lineage tracking (Iceberg V3-style) with monotonic row IDs, mutation versioning, and SHA-256 per-row hash chain for tamper evidence.
	streaming_sink.hpp
	Lock-free SPSC/MPSC ring buffers, StreamingSink for background Parquet compaction, and HybridReader for querying across historical and live data.
	tensor_bridge.hpp
	Zero-copy tensor bridge: maps Parquet column data directly into ML-framework-compatible tensor views (ONNX Runtime, PyTorch, etc.) without copying.
	threat_model.hpp
	vector_type.hpp
	AI-native ML vector column type: VectorWriter/VectorReader for dense float embedding storage as FIXED_LEN_BYTE_ARRAY Parquet columns, plus SIMD-accelerated dot product, L2 distance, normalization, and copy.
	wal.hpp
	Write-Ahead Log (WAL) with sub-millisecond append, CRC-32 integrity, crash recovery, and automatic segment rolling.
	wal_mapped_segment.hpp
	Cross-platform memory-mapped WAL segment and ring writer.