My Projects

Châu Thảo and a teammate holding their project awards
Châu Thảo beside the architectural AI research poster
The team presenting their AI project in a classroom
Châu Thảo participating in a research discussion

Innovation 1: Road Monitor

Smart Subsidence Detection and Analysis for Greener Urban Mobility

Co-Programmer | AI Challenge Contest – EcoGen 2026 | June – July 2026

I collaborated with a five-member team to develop Road Monitor, an AI-powered system that detects road potholes and explores their impact on safer and more sustainable urban mobility. I contributed to the AI research and development of the detection system using YOLO26n.

I trained and evaluated the model on 665 public-dataset images, achieving 0.77 Precision, 0.73 Recall, and a 0.75 F1-score for pothole detection. I also contributed to developing real-time pothole alerts and eco-routing concepts; an HCMC test involving 20 motorbikes recorded 38% lower idling and 22% lower fuel consumption, demonstrating the potential of AI to support greener urban transportation.

Github

Innovation 2

Generative AI for Architectural Style Transformation | International Journal of Engineering and Information Systems

Co-Programmer | IJEAIS | April – August 2026

I co-developed ArchStyleAI, a generative AI pipeline that combines Stable Diffusion v1.5, ControlNet with M-LSD line maps, and LoRA style adapters to transform architectural styles while preserving key structural elements such as columns and window grids.

I evaluated the system across different Img2Img Strength settings using 50 Gothic facade images, measuring both visual quality and structural fidelity through CLIP Score, LPIPS, and a line-map SSIM metric. The experiments identified S = 0.75 as the optimal setting for balancing stylistic transformation with architectural integrity.

Github

Research

Generative AI for Architectural Style Transformation

Researcher | International Journal of Engineering and Information Systems (IJEAIS) | April – August 2026

I conducted a 5-month, 5,000-word research study investigating how Generative AI can transform architectural styles while preserving the structural identity of buildings.

I experimented with CycleGAN and Stable Diffusion to transform architectural designs across styles, including Modern to Gothic and Traditional to Futuristic. Through comparative evaluation, I examined the trade-off between stylistic transformation and structural preservation, developing research experience in generative AI, computer vision, image-to-image transformation, and AI-generated content evaluation.