konkovdv/aerobim

Purpose and entry points

AeroBIM is an openBIM acceptance-criteria assistant for IFC, IDS, drawings, requirements, and calculation evidence. It produces a deterministic ISO-19650 Shared-gate report with provenance, capability status, exports, and optional advisory AI; it is explicitly not a customer-ready CDE/MEP/calculation-correctness product.

Backend starts at backend/src/aerobim/main.py; FastAPI routing is assembled in backend/src/aerobim/presentation/http/api.py. The browser review shell starts at frontend/src/main.tsx and frontend/src/App.tsx.

Layered backend and composition

The Python backend follows inward Clean Architecture: domain models/ports, application use cases, infrastructure adapters, then HTTP presentation. Constructor DI binds ports to enabled adapters and environment-selected implementations.

• backend/src/aerobim/domain/ports.py

• backend/src/aerobim/core/di/container.py

• backend/src/aerobim/infrastructure/di/bootstrap.py

Configuration determines fail-closed profiles, authorization, optional engines, storage, and AI routing through backend/src/aerobim/core/config/settings.py.

Package ingestion and analysis orchestration

The central flow accepts uploaded/stored package inputs or direct analysis requests, constructs a package trace, applies completeness/intake gates, and runs the project-package orchestration. Larger requests can use the same-process job submission/polling path.

• backend/src/aerobim/presentation/http/routes/uploads.py

• backend/src/aerobim/application/use_cases/analyze_project_package.py

• backend/src/aerobim/application/use_cases/analyze_project_package_jobs.py

Outputs aggregate findings, evidence references, run manifest, capabilities, and a PackageOutcome; failed required capabilities prevent a passing Shared-gate result.

Deterministic validation engines

The deterministic core fuses IFC property/quantity checks, IDS validation, requirements extraction/interpretation, cross-document contradiction detection, drawing annotations, and optional clash/MEP checks.

• backend/src/aerobim/application/services/analyze_orchestrators.py

• backend/src/aerobim/infrastructure/adapters/ifc_open_shell_validator.py

• backend/src/aerobim/infrastructure/adapters/ifc_tester_ids_validator.py

Requirements can be compiled to IDS via backend/src/aerobim/application/use_cases/compile_requirements_to_ids.py; adapters isolate IfcOpenShell, IfcTester, Docling, CAD/PDF, and optional geometry dependencies from domain policy.

Evidence, verdict, and human review boundary

Domain objects encode findings, provenance, capability contracts, review state, signatures, and reproducibility. Verdict ownership is deterministic; advisory and HITL events do not independently turn a report into a pass.

• backend/src/aerobim/domain/findings.py

• backend/src/aerobim/domain/finding_provenance.py

• backend/src/aerobim/application/use_cases/apply_norm_rule_hitl_event.py

Reports support list/detail, review-event, KPI, and source/preview retrieval routes in backend/src/aerobim/presentation/http/routes/reports.py.

AI and drawing advisory boundary

Hybrid routing classifies data and applies privacy/trust policy before optional LLM/VLM advisory calls. It is verdict-neutral (summary.passed must be invariant with AI off/on), budgeted, audited, and guarded against unsafe egress.

• backend/src/aerobim/application/services/hybrid_route_gate.py

• backend/src/aerobim/domain/hybrid/model_router.py

• backend/src/aerobim/infrastructure/adapters/openai_compat_llm_provider.py

Drawing analysis uses structured/raster/region-restricted pipelines and evidence grounding, but human-level CV remains an explicitly missing capability.

Persistence, exports, and integrations

Report payloads and binary artifacts sit behind ports: local filesystem/object storage by default, optional S3/MinIO artifacts and Postgres summary indexing. BCF 2.1/3.0 ZIP is exported structurally; API push is experimental and independent CDE import is not verified.

• backend/src/aerobim/infrastructure/adapters/local_object_store.py

• backend/src/aerobim/infrastructure/adapters/s3_object_store.py

• backend/src/aerobim/infrastructure/adapters/bcf3_exporter.py

Relevant HTTP export endpoints are in backend/src/aerobim/presentation/http/routes/exports.py; BCF API push is represented by backend/src/aerobim/application/use_cases/push_report_to_bcf_api.py.

Runtime and client surfaces

Material runtime: Python 3.12+, FastAPI/Uvicorn in Docker; compose definitions are docker-compose.yml and docker-compose.production.yml. Optional infrastructure includes Redis job storage, Postgres, and S3-compatible storage. The TypeScript/Vite frontend renders IFC via web-ifc/Three.js and presents findings, provenance, capability honesty, and drawing evidence.

• frontend/src/lib/api.ts

• frontend/src/lib/ifc-scene.ts

• frontend/src/components/ProvenancePanel.tsx