Quesarby MLAI

Research · implementation

A six-layer architecture for evidence-aware AI systems

Quesar organizes interface, orchestration, retrieval, acceleration, storage, and audit as distinct layers so that evidence and runtime claims can be traced to the component that actually supports them.

AIWDBXSEAGPUMCPTUI

Architecture

The platform model places user-facing interfaces at L6, orchestration at L5, retrieval and learning at L4, acceleration at L3, durable storage at L2, and provenance and audit at L1. The separation is practical: an MCP or TUI surface can coordinate work without being mistaken for the storage engine, and a GPU path can accelerate computation without becoming the source of record.

The accompanying platform catalog maps concrete MLAI systems onto those layers. ABI spans orchestration and interface concerns, WDBX supplies storage and retrieval primitives, SEA contributes evidence-sensitive learning, GPU backends are acceleration paths, and MCP and TUI expose bounded control surfaces.

Evidence discipline

The catalog records capability limits alongside descriptions. CPU execution is treated as the correctness oracle for accelerated paths, cluster RPC is not presented as automatic distributed sharding, and WebGPU or TPU entries remain capability descriptions until their specific runtime evidence is available.

This turns the layer diagram into a claim boundary rather than a feature collage: each public statement should identify its implementation layer, evidence artifact, and unsupported extension.

Limitations

  • The layer model is an architectural taxonomy, not evidence that every mapped subsystem is deployed together in one production environment.
  • Acceleration, clustering, and security claims still require backend-specific tests and operational evidence; adjacency in the stack does not establish those properties.
All implementationsA private document pipeline with explicit fallback behavior
Research index

Next

AI overview

The track note this case sits under.