Agentic-BIM-Compliance

An autonomous, Agentic AI Co-Pilot designed to automate structural engineering design reviews. By parsing Industry Foundation Classes (IFC) models against Indian Standard codes (IS 456, IS 1893, IS 800), the system compresses a typical 2-day manual review cycle into just 4 minutes.

It provides structural engineers with explainable AI outputs, interactive 3D visualizations, and an automated remediation loop for non-compliant members.

---

StructurAI Co-Pilot utilizes a highly modular, multi-agent architecture orchestrated by LangGraph. It seamlessly integrates deep IFC parsing, Vision-guided RAG, and an immutable audit trail.

1. Input Layer: Ingests IFC building models alongside structured SP 16 tables and IS/BIS Code PDFs (IS 456, 800, 1893, NBC 2016).
2. Parsing & Retrieval: - IfcOpenShell performs deep parsing to extract axis, topology, and PropertySets.

• Vision RAG + HyDE handles complex code clause retrieval using BM25 and GraphRAG for cross-document reasoning.

1. Agentic Core: A Supervisor Agent (LangGraph State Machine) acts as the central brain, coordinating the IFC parser, code retriever, mathematical calculation engine, and MCP tools.
2. Output & Visualization: Renders the model using xeokit 3D viewer with BCF fly-to capabilities, color-coded safety bars, and ranked failures.
3. Auto-Remediation Loop: Proposes fixes, calculates the embodied carbon delta, and provides a Human-In-The-Loop (HITL) diff view before re-evaluating the model.
4. Audit & Chat: All decisions