🚧 MOD Code Jam Challenge: BIM Context Graph + Agentic Assistant

Welcome to the MOD AI Code Jam! This is a collaborative and creative challenge where we'll prototype a core subsystem for MOD’s future Contextual GNN — with an agentic twist.

---

🧠 Goal

Build a modular pipeline that:

1. Parses a BIM model from IFC format
2. Extracts contextual relationships (e.g., adjacency, containment, connectivity)
3. Constructs a graph abstraction (nodes + edges + attributes)
4. Wraps this into a lightweight agent that responds to user prompts like:

   “Build a graph with only load-bearing walls and adjacent doors.”

---

🎯 Why This Matters

MOD’s roadmap includes a Contextual GNN Model (Q4 2025) that interprets spatial logic in BIM models to support prefab validation and optimization. This challenge kicks off that direction.

You’ll also simulate how this capability would be embedded in a LangGraph-style agent, part of MOD’s broader AI-native architecture.

---

👥 Team Roles

• Lukas:
  Lead BIM logic, define geometry extraction from IFC using tools like ifcopenshell.

• Ata & Emilio:
  Lead graph extraction logic, feature engineering, and agent construction.

• Yashar:
  Facilitate system modularity, ensure extensibility, and connect the work to MOD's GNN and mod-val/mod-rec future.

---

🔧 Tasks Overview

1. Parse IFC Model

• Use ifcopenshell (Python) to extract relevant spatial elements from a provided .ifc file
• Transform those elements into structured JSON (e.g., rooms, walls, doors with metadata)

2. Build Context Graph

• Nodes: prefab-suitable elements (e.g., rooms, walls, doors)
• Edges: spatial/semantic relationships
  • adjacent_to
  • contained_in
  • connected_to
• Attributes: type, dimensions, load-bearing status, etc.
• Output: Graph object (e.g., NetworkX) + JSON

3. Add Agentic Wrapper

Build a lightweight agent that:

• Accepts natural-language prompts
• Parses intent (e.g., which types of nodes/edges to include)
• Filters input and calls the graph builder
• Returns:
  • Graph (JSON or visualization)
  • Short explanation (e.g., “Graph has 6 nodes, 3 adjacency edges.”)

Bonus: Use GPT for prompt parsing or post-hoc explanation.

---

⏱️ Suggested Time Breakdown

Phase	Duration	Focus
Intro	30 min	Walkthrough of challenge + Q&A
Coding	2 hrs	Build IFC parser + graph + agent
Debrief	30 min	Share solutions and discuss takeaways

---

📦 Inputs

We'll provide:

• A sample .ifc file representing a simple BIM scenario
• This repository includes optional starter utilities, type definitions, and graph I/O functions to help you get started quickly.
• Option to use OpenRouter or local GPT (via Ollama) if desired

---

📤 Output

• Python script or notebook that:
  • Parses IFC → structured JSON
  • Builds a context graph
  • Supports filtered prompt input via agent
• Graph output in JSON or networkx format
• (Optional) Visualization or GPT-based explanation
• Brief README or summary of approach

---

🔄 Processing Pipeline

graph TD
    A["IFC File (.ifc)"] --> B["IFC Parser (parse_ifc.py)"]
    B --> C["Structured JSON of BIM Elements"]
    C --> D["Contextual Graph Builder (graph_builder.py)"]
    D --> E["Agentic Assistant (agent.py)"]
    E --> F["Graph Output + Summary"]

Loading

---

🧑‍🔧 Tech Stack Suggestions

• Language: Python
• Libraries: ifcopenshell, networkx, json, pydantic, trimesh, LangChain (optional)
• GPT Access: OpenRouter or local Ollama (if used)
• Agent Emulation: Minimal planning → tool call → return

---

🗂️ Project Structure

ai-code-jam-gnn/
├── data/
│   └── sample.ifc              # Sample BIM model in IFC format
├── src/
│   ├── parse_ifc.py               # IFC → JSON extraction script
│   ├── graph_builder.py           # Core graph construction logic
│   ├── agent.py                   # Agent interface for prompt interpretation
│   ├── main.py                    # Entry point to run full pipeline
│   └── utils.py                   # Helper functions (optional)
├── requirements.txt               # Dependencies used in the project
└── README.md

Architecture Solution

IFC Parsing

• walls
• IfcWall
• slabs
• spaces

Metadata

• Name
• Guid Geometry
• Boundingbox

Remark: Validate if all Types are included within groups wall,slab,room

Connectivity

Test topologicpy to create graph

• adjacent_to
• contained_in

Agentic Wrapper

🚀 Let’s Build the First MOD Graph Tool!

This challenge is an entry point into the graph abstraction layer that will power MOD's contextual reasoning, validation, and prefab recommendations. It's also a test of how domain expertise, geometric logic, and AI orchestration come together.

Make it modular. Make it explainable. And have fun with it!