GARC Case Study. A Full Modern Architecture for Computational Design
GARC is GATAI’s unified, multimodel platform for computational design.
It transforms a simple text brief into validated outputs across the pipelinefrom Floor
Plan Generation (Beta) to upcoming modules for materials and heat insulation,
physics-aware constraints, and 3D export.
The system is built with a cloud-native architecture:
modular services behind an API gateway, an event bus with queued workers, real-time
progress streaming, and a secure data layer for plans, materials, and 3D assets.
The Problem
Design and construction teams today rely on disconnected tools:
One model for plan sketches
A separate database for materials and thermal properties
Another engine for physics checks
Ad-hoc scripts for 3D exports
This fragmentation slows down iteration, increases handoff risks, and makes scaling across teams costly.
Goals
Unify capabilities behind a single API and dashboard
Make long-running jobs observable with progress tracking
Keep deployment flexible across cloud, on-prem, and hybrid setups
Enforce strong security and auditability with RBAC and logs
Architecture at a Glance
GARC is organized into modular services that communicate over an event bus:
API Gateway versioned endpoints, authentication, rate limits
Services Floor Plans (Beta), Materials & Thermal, Physics, 3D Export
Orchestrator job scheduling, GPU allocation, retries, and progress events
Data Layer storage for plans, metadata, and assets with full audit trails
Observability structured logs, metrics, and job timelines
Floor Plan Generator (Beta)
The Floor Plan module turns a text prompt into draft layouts.
Workflow:
Submit a prompt with parameters (e.g. number of images, scheduler, inference steps).
Receive a task ID and observe progress updates in real time.
Fetch final results (PNG URLs or data URIs) via the status endpoint.
This same asynchronous pattern will power future modules, so product teams can use one consistent interface for all GARC services.
End-to-End Orchestration
GARC pipelines let teams chain tasks seamlessly:
Text to Floor Plan (Beta)
Materials & Thermal analysis (coming soon)
Physics validation (coming soon)
3D Export into GLTF/OBJ formats (coming soon)
The orchestrator ensures dependencies are respected, retries failed jobs, and keeps clients updated throughout.
Data & Security
RBAC & Audit Logs for organizational accountability
Secrets Management to isolate environments securely
Metadata Search across plans, textures, and 3D assets
Content-Addressed Storage for reproducibility and deduplication
Deployment Options
Cloud autoscaling GPU pools for heavy loads
On-Premise secure, air-gapped deployment for regulated industries
Hybrid sensitive data local with scalable cloud compute for burst workloads
Impact
Early adopters highlight:
Time-to-first-plan reduced from hours to minutes
Unified job lifecycle and observability across modules
Cleaner architectural boundaries, allowing each service to evolve independently
Roadmap
Materials & Heat Isolators thermal analysis per room and climate
Physics Checks circulation, clearances, and constraint validation
3D Builder / Architect white-box layouts with asset placement
SDKs expanded client libraries and typed stubs for developers
