Agentic LLM (Qwen3) for weld defect detection in industrial radiographic image
Here we build a simple multi-system prompts that used Qwen3-VL-8B-Instruct-Q4_K_M.gguf and mmproj-F16.gguf with lovely results without fine-tuning the model.
Tuning were only done on the multi-system prompts but the images used for testing were CLAHE improved. The inference file can be used to detect and classify
weld defects in industrial radiographic image.
This development is part of effort at NDT-Material Structural Integrity Group at Agensi Nuklear Malaysia to build
agentic AI for NDT inspection.
Simple inference file: LLM_Inference.ipynb
System prompt files (decision tree agents): prompt_router_3.txt, prompt_volumetric_3.txt, prompt_planar_3.txt
All these files (including the model) will be updated from time to time when better results and system are achieved.
