Forge Sight Quality Control

Vercel
application badge
Created by team Ras Ali Labs on May 04, 2026
AI Agents & Agentic Workflows (Best Track for Beginners)Fine-Tuning on AMD GPUs (Advanced / GPU-Intensive)Vision & Multimodal AIHugging FaceQwen

ForgeSight is a multimodal quality control copilot specifically designed for manufacturing environments. It automates the defect detection and reporting process through a unique four-agent pipeline that includes an Inspector, Diagnostician, Action, and Reporter. These agents collaborate to inspect assembly line images, diagnose the root causes of defects, draft work orders, and publish fully auditable reports. Developed for the AMD Developer Hackathon, the system utilizes a fine-tuned Qwen2-VL vision model hosted on an AMD Instinct MI300X node, employing ROCm and vLLM. By leveraging the impressive 192 GB of memory available on a single MI300X GPU, the 72-billion parameter model operates efficiently without the need for sharding. This capability allows it to deliver real-time inference on the factory floor and provide verdicts in just 20 seconds.

Category tags: