Abstract
A VR network visualization system using voice commands and LLM-powered agents to transform natural language into Neo4j queries, enabling interactive graph exploration through categorical coloring, shape encoding, and dynamic layouts.
Skills
Awards
ISMAR 2026 Paper Submission (Under Review)
Link
Contents
This is a project with one lead PhD and two main developers including me. I designed and implemented the multi-agent LangGraph pipeline that transforms voice input into visualization operations, including speech error correction and ambiguity detection. Also, I am responsible for the categorical encoding system for bivariate visualization (color and shape by attribute) with automatic mapping and legend generation. I also created the end-to-end execution flow integrating Whisper STT, LangGraph, Neo4j, and Unity rendering.
This real-time visualization systems that help people understand complex data through interactive and intuitive interfaces. My work focuses on the intersection of graphics, system performance, and user experience. I developed a VR-based network visualization system in Unity, where I designed and implemented a GPU-accelerated rendering pipeline to handle large-scale graph data. I used techniques such as compute shaders for parallel curve generation and GPU instancing to efficiently render thousands of elements in real time.
