Auraison Platform

What is Auraison?

Auraison is an agentic AI platform that orchestrates GPU workloads, experiment tracking, and data pipelines through Claude Code subprocess agents. It follows a four-plane architecture (user, control, data, management) inspired by SDN/telecom separation patterns.

The platform provides the infrastructure on which reference applications run. The platform itself is application-agnostic — it manages clusters, jobs, data, and agent lifecycles.

Platform Architecture

Plane	Responsibility	Key Components
Control plane	Job dispatch, agent lifecycle, experiment tracking	FastAPI API, AgentOps, Claude Code agents, MAC framework
User plane	Workload execution on KubeRay	torch.dev.gpu, ros.dev.gpu Ray environments
Data plane	Persistent storage, context retrieval	DuckDB + DuckLake + MinIO lakehouse
Management plane	Billing, tenancy, quotas	v2 placeholder

See architecture/four-plane.md for the full design.

MAC: Multi-Agent Control

The MAC framework (control-plane/backend/mac/) provides the information-theoretic foundation for multi-agent communication — treating agent outputs as noisy channel observations and applying coding-theoretic techniques (majority vote, soft consensus, turbo iterative fusion, Viterbi decoding) to reduce hallucination. See mac-principles.md for the theoretical framework.

Reference Applications

Auraison supports three reference applications. Each is a domain-specific system deployed onto the platform's KubeRay clusters.

Application	Domain	Repository	User Plane	Status
turtlebot-maze	Mobile robot navigation	aegean-ai/turtlebot-maze	ros.dev.gpu	Active
ar4-physical-ai	Robotic arm manipulation (VLA)	aegean-ai/ar4-physical-ai	ros.dev.gpu + torch.dev.gpu	Active
Deep Evidence Agent (DEA)	Engineering traceability & evidence-grounded reasoning	aegean-ai/dea	torch.dev.gpu	Design phase

Deep Evidence Agent (DEA)

The DEA is a multi-agent system for safety- and mission-critical engineering organizations (telecom, aerospace, automotive, medical devices). It turns scattered engineering artifacts (requirements, design docs, code, tests, standards) into a traceable, auditable knowledge base with evidence-grounded reasoning. Uses Microsoft GraphRAG for graph-based retrieval-augmented generation.

• Repository: aegean-ai/dea (includes graphrag as git submodule)
• Docs: PRD, ADD, SDD in the dea repo under docs/

Documentation Structure

docs/
├── architecture/         Platform-level architecture (four-plane)
├── control-plane/        Control plane design docs + MAC principles
├── user-plane/           User plane design, digital twins, reference app designs
├── data-plane/           Data plane design, lakehouse demos
├── management-plane/     Management plane design (v2)
└── devops/               Revision history, sprint plans