Byte Bandits

A.I.R.E.N is an adaptive 3D CAD intelligence platform that solves a core frustration: traditional CAD tools are slow, rigid, and require heavy manual operations. A.I.R.E.N replaces this with an agent-driven design system that understands spatial intent, interprets natural inputs, and generates real-time parametric models. The user describes or gestures a structure, the agent translates it into geometry, refines constraints, and produces an editable 3D model. The end-to-end demo shows input → interpretation → model generation → refinement in one continuous flow.

---

How the Project Meets Judging Criteria

1. Working Prototype

A functioning pipeline where user prompts (voice/text/gesture) are parsed by an AI agent, converted into geometric instructions, and rendered as 3D models using a live viewer. Real-time edits demonstrate responsiveness.

2. Technical Complexity & Integration

Multi-agent system handles interpretation, geometry planning, constraint solving, and visualization. Integrates LLM reasoning, vector embeddings, parametric modeling logic, and WebGL rendering. Demonstrates continuous loop between model generation and user intent correction.

3. Innovation & Creativity

Reimagines CAD as an intelligent collaborator instead of a tool. Natural-language-driven modeling and predictive geometry are rarely achieved in lightweight prototypes. Spatial reasoning + AI agent orchestration distinguishes it.

4. Real-World Impact

Enables designers, engineers, students, and makers to create complex models faster with minimal learning curve. Reduces the barrier to 3D design and accelerates prototyping in manufacturing, architecture, robotics, and product design.

5. Theme Alignment

Built around an AI agent that autonomously interprets instructions, makes decisions, and executes design tasks end-to-end. Fully demonstrates agentic reasoning, multi-step planning, and adaptive correction.

---

Technologies & Integrations Used

Core AI & Agents
• GPT-based LLM for spatial reasoning
• Embedding model for intent mapping
• Agent orchestration via LangChain/AutoGen-style frameworks
• Custom geometry-planning agent (constraints, dimensions, topology)

Rendering & Modeling Stack
• Three.js / Babylon.js for WebGL 3D visualization
• OpenCascade / CADKernel for parametric modeling operations
• Python backend for geometry computation
• REST/WebSocket layer for real-time sync

Frontend & Interaction
• React + TypeScript
• Voice input (Web Speech API)
• Optional gesture/mouse-driven sketch mode

Infrastructure
• Cloud deployment via Vercel/Render
• Python backend hosted on Railway/Fly.io
• Storage for model files (S3-compatible buckets)

---

Demo & Reproducibility Steps

1. User enters prompt or voice command: “Create a cylindrical base with a 10 cm radius and attach a hollow cube on top.”

2. Agent parses prompt → constructs intent map → generates parametric instructions.

3. Geometry engine constructs the model and renders it instantly in the browser.

4. User asks for refinements (“Make the walls thinner” or “Rotate the cube 30 degrees”).

5. Agent updates constraints and re-renders in real time.

6. Final model can be exported as STL/STEP.