5-layer pipeline deep dive
Complete internal architecture of Aucert's 5-layer AI quality pipeline with MCP protocol, code paths, and model assignments
Complete internal architecture of Aucert's 5-layer AI quality pipeline with MCP protocol, code paths, and model assignments
Context
Context
Context
Context
Context
Context
Context
Context
Context
Context
Substantially revised 2026-06-20: adds the by-domain folder layout, the three-tier contract model,
Companion to ADR-020 (which covers the
Context
Why Aucert has no LLM models running in AKS — and where they actually run
Internal architecture of the AI Device Twin — calibration, prediction, and confidence adjustment for emulator-to-device divergence
Internal architecture of the Aucert Knowledge Graph engine — data model, storage, ingestion pipeline, and query patterns
How Aucert routes AI workloads across model tiers for cost optimization and quality balancing
Interactive walkthrough of the Validation Graph design (SPEC-035) — two-graph knowledge layer, substrate, ACL, conditions, embeddings, deployment, tech choice
How the merged validation graph (SPEC-035) works in code — the append-only claim model, the storage engine's read/write/traversal/bitemporal paths, registries, ACL, vector search, and the cross-cutting RLS, DI, and open-core patterns.
4-stage confidence-gated verification system for minimizing false positives with cost-aware escalation